AN IMPROVED NON-LOCAL DENOISING ALGORITHM

Size: px
Start display at page:

Download "AN IMPROVED NON-LOCAL DENOISING ALGORITHM"

Transcription

1 AN IMPROVED NON-LOCAL DENOISING ALGORITHM Bart Goossens, Hiêp Luong, Aleksandra Pižurica and Wilfried Pilips Gent University, Department of Telecommunications and Information Processing (TELIN-IPI-IBBT Sint-Pietersnieuwstraat 4, B-9000 Gent, Belgium ABSTRACT Recently, te NLMeans filter as been proposed by Buades et al. for te suppression of wite Gaussian noise. Tis filter exploits te repetitive caracter of structures in an image, unlike conventional denoising algoritms, wic typically operate in a local neigbourood. Even toug te metod is quite intuitive and potentially very powerful, te PSNR and visual results are somewat inferior to oter recent state-of-te-art non-local algoritms, like KSVD and BM-3D. In tis paper, we sow tat te NLMeans algoritm is basically te first iteration of te Jacobi optimization algoritm for robustly estimating te noise-free image. Based on tis insigt, we present additional improvements to te NLMeans algoritm and also an extension to noise reduction of coloured (correlated noise. For wite noise, PSNR results sow tat te proposed metod is very competitive wit te BM-3D metod, wile te visual quality of our metod is better due to te lower presence of artifacts. For correlated noise on te oter and, we obtain a significant improvement in denoising performance compared to recent wavelet-based tecniques.. INTRODUCTION Digital imaging devices inevitably produce noise, originating from te analog circuitry in tese devices. Noise suppression by means of digital post-processing is often desirable, but also very callenging. In tis paper, we focus on te design of a noise reduction metod for stationary Gaussian noise, tat preserves original image details and tat as a ig visual quality. During te past decade, numerous and diverse denoising metods ave been proposed to tis end. Many metods, like total variation [], bilateral filtering [] or wavelet-based tecniques [3 9] estimate te denoised pixel intensities based on te information provided in a limited surrounding neigbourood. Tese metods only exploit te spatial redundancy in a local neigbourood and are terefore referred to as local metods. Recently, a number of non-local metods ave been developed. Tese metods estimate every pixel intensity based on information from te wole image tereby exploiting te presence of similar patterns and features in A. Pižurica is a postdoctoral researcer of te Fund for te Scientific Researc in Flanders (FWO Belgium. an image. Tis relatively new class of denoising metods originates from te Non-Local Means (NLMeans, introduced by Buades at al. [0, ]. Basically, te NLMeans filter estimates a noise-free pixel intensity as a weigted average of all pixel intensities in te image, and te weigts are proportional to te similarity between te local neigbourood of te pixel being processed and local neigbouroods of surrounding pixels. Oter non-local denoising metods are exemplar-based (KSVD [], or group similar blocks by block-matcing and ten apply 3D transform-domain filtering to te obtained stacks (BM-3D [3]. Te NLMeans filter, despite being intuitive and potentially very powerful, as two limitations at tis moment: first, bot te objective quality and visual quality are somewat inferior to te oter recent non-local tecniques and second, te NLMeans filter as a complexity tat is quadratic in te number of pixels in te image, wic makes te tecnique computationally intensive and even impractical in real applications. For tis reason, improvements for enancing te visual quality and for reducing te computation time ave been proposed by different researcers. Some autors investigate better similarity measures [4 6], use adaptive local neigbouroods [7], or refine te similarity estimates in different iterations [8]. Oter autors propose algoritmic acceleration tecniques [6,9 ], based for example on neigbourood preclassification [6,9] and FFT-based computation of te neigbourood similarities [0]. In tis paper, we sow te connection between te NLMeans filter and robust estimation tecniques, similar to te connection made for te bilateral filter in []. It turns out tat te NLMeans filter is te first iteration of te Jacobi algoritm [] (also known as te Diagonal Normalized Steepest Descent algoritm for robustly estimating te noise-free image using te Leclerc loss function. By tis observation, it becomes possible to investigate oter robust cost functions tat are commonly used for robust estimation. Also, tis suggests tat applying te NLMeans algoritm iteratively in a specific way, furter decreases te cost function. Anoter problem noted by Buades et al. is tat te NLMeans filter is not able to suppress any noise for non-repetitive neigbouroods. In tis work, we keep track of te local noise variance during NLMeans filtering and we apply a local post-processing

2 filter afterwards, to remove remaining noise in regions wit non-repetitive structures. Furtermore, we present an extension of te NLMeans filter to correlated noise and we also present a new acceleration tecnique tat computes te Euclidean distance by a recursive moving average filter. Te proposed modifications significantly improve te NLMeans filter, bot in PSNR as in computation time, and make te filter competitive to (or even better tan recent non-local metods suc as BM-3D of Dabov et al. [3]. Te remainder of tis article is as follows: in Section, te NLMeans algoritm is briefly presented. In Section 3, we investigate te coice of te similarity weigting functions, on probabilistic reasoning and witin te robust estimation framework. In Section 4, we present our improvements to te NLMeans algoritm. We discuss ow te computation time of te filter can be furter reduced in Section 5. Numerical results and visual results togeter wit a discussion, are given in Section 6. Finally, Section 7 concludes tis paper.. NON LOCAL MEANS Suppose an unknown signal X i,i=,..., N is corrupted by an additive noise process V i,i=,..., N, wicresults in te observed signal: Y i = X i + V i ( In tis work, we stick to definitions for -d signals for simplicity of te notations. An extension to -d images is straigtforward. Te denoised value ˆX i of te pixel intensity at position j is computedas te weigted average of all pixels in te image: ˆX i = N j= w(i, jy j N j= w(i, j ( We will refer to tis filter as te pixel-based NLMeans. Alternatively, a vector (or block-based NLMeans filter does also exist [0]. Here overlapping blocks are used, resulting in multiple estimates for eac pixel in a block. To aggregate te different estimates, an additional weigting function b(k determines te weigt contribution of central pixel to its neigbour at relative position k: N K j= k= K ˆX i = b(kw(i + k, j + ky j N K j= k= K b(kw(i + k, j + k (3 Te weigts w(i, j depend on te similarity between te neigbouroods centered at positions i and j. In tis paper, neigbouroods of fixed predefined size are used (e.g Let us denote x i = (X i K,...,X i+k, y i =(Y i K,...,Y i+k and v i =(V i K,...,V i+k as vectors containing pixel intensities of te local neigbourood centered at position i (were for example symmetrical boundary reflection is used at te signal/image boundaries. In te original NLMeans algoritm [0, ], te weigting function is defined as follows: ( w(i, j =exp y i y j (4 wit y i y j = (y i y j T (y i y j te Euclidean distance between te vectors y i and y j. Te bilateral filter [] is closely related to te NLMeans filter. For te bilateral filter, te weigting function is given by: ( w(i, j =exp (Y i Y j exp ( (i j d (5 were te first factor (called potometric distance is inversely proportional to te Euclidean distance between te pixel intensities Y i and Y j and te second factor (called geometric distance, measures te Euclidean distance between te center sample i and te j-t sample. Te NLMeans weigting function can be interpreted as a vector-extension of te bilateral filter weigting function, omitting a geometric distance factor. In [, ], te weigting functions as given in equations (5 and (4 are defined on intuition. However it is not guaranteed tat tese coices are optimal for a given criterion. In [], it is sown tat equation (5 emerges from optimizing a robust M-function. In te next Section, we will derive expressions for w(i, j based on probabilistic reasoning and we will furter extend te result from [] to te NLMeans filter. 3. COMPUTING SIMILARITY WEIGHTS One of te key elements for designing a ig performance NLMeans filter, is te selection of te similarity weigts. Clearly, te similarity weigts sould be adapted to te image in order to acieve maximal improvement. Because only a noisy version of te image is available, te weigts sould also take te noise properties into account; te presence of noise generally degrades te estimate of te similarity between two neigbouroods in te image. Te question now becomes: ow to determine te similarity weigts in an optimal sense for a givencriterion? Luckily, estimation teory can give an answer to tis problem. 3.. Best Linear Unbiased Estimator (BLUE We now only consider te estimation of te noise-free patc x i as a function of surrounding noisy patces. To take te correlation between different x i into account, we model te residual r i,j between patces (centered at position i and j as: x i = x j + r i,j (6 wit r i,j a zero-mean Gaussian random variable wit position-dependent covariance matrix σi,j I (te disadvantages of tis coice will be discussed furter on and wit σ i,i =0. Due to te additivity of te noise, we ave: y i = x j + r i,j + v i (7 For wite Gaussian noise, v i N(0,σ w I. Te Best Linear Unbiased Estimator (BLUE for tis problem is given

3 by: ˆx i = argmin x = argmin x = N j= N j= N log f y (y j ; x (8 j= N j= σ w +σ i,j y j σ w +σ i,j y j x σw + σ i,j (9 (0 Te minimum can be found iteratively, e.g. by gradient descent. Applying only one iteration of tis algoritm yields: ˆx i = y i λ i N j= ρ (y i y j (4 wit ρ (x te gradient of ρ(x. An interesting case is obtained by considering a robust function wit derivative of te form: ρ (x =xg(x, In tis case, equation (4 becomes: wic ere also corresponds to te maximum likeliood (ML estimator. Te variances σi,j,i j are unknown and need also to be estimated. Noting tat from: (y j y i =(r j,j r i,j +(v j v i it follows tat Var [y j y i ]= ( σ w +σ i,j I,tefollowing estimate can be obtained: σ i,j =max (0, (K + y i y j σw ( Tis solution suggests te use of te following weigt matrix: 4K+ y i y j y w(i, j = i y j 4K+ >σw y i y j ( 4K+ σw σ w Tis means tat for i j te weigts are inversely proportional to te Euclidean distance between te vectors at position i and j. Unfortunately, te Gaussian distribution for modeling te residuals is not a good coice: for natural images we may expect tat for eac vector x i tere are many oter vectors x j tat are very similar to x i. Te residual or difference between x j and x i would from tis perspective rater ave a Laplacian distribution tan a Gaussian distribution. Moreover, dissimilar patces x i tend to cluster, wic causes multiple modes in te istogram of te residual. One possibility is e.g. to use a better suited multivariate (multimodal distribution for r i,j tat also incorporates correlations between te components of r i,j. Because te estimators for suc a distribution are muc more complicated and te model training is computationally more intensive, an attractive alternative is to use robust statistics instead and to treat te deviations from te model as outliers. 3.. Robust M-estimator Te Robust M-estimator is derived from te ML estimator by replacing te quadratic term in te negative loglikeliood functional by a multivariate robust loss function ρ(x, as follows: ˆx i =argmin x N ρ (x y j (3 j= ˆx i = y i λ i N = λ i λ i j=,j i N j=,j i N j=,j i g (y i y j (y i y j g (y i y j y i + g (y i y j y j (5 To speed up te first iteration, te step-size λ i can be adapted based on te Jacobi algoritm. Analogous to [], tis leads to: λ i = Equation (5 becomes: + N j=,j i g (y i y j ˆx i = y i + N j=,j i g (y i y j y j + N j=,j i g (y i y j (6 (7 wic is te NLMeans filter wit weigt function given by: { g (y i y j i j w(i, j = (8 i = j Example: for te Leclerc robust function defined by: ( ρ(r = exp rt r (9 we ave: and ( ρ (r =r exp rt r ( w(i, j =exp y i y j (0 ( Hence, te Leclerc robust function leads us to te weigting function from equation (4. Interestingly, te weigt function ( derived in Section (3. can also be interpreted to be associated to a robust function: ρ(r = { r +log r r > r (

4 Tis interpretation makes it easier to compare te properties of te resulting estimator wit oter robust estimators. Note tat te robust estimation framework gives us a muc wider variety of weigting functions. Different robust functions will be compared more in detail in Section IMPROVEMENTS TO THE NLMEANS FILTER Based on te teoretical framework explained in Section and Section 3, we are now able to present a number of improvements for te NLMeans algoritm: If only one iteration of te NLMeans filter is applied, te solution as generally not converged to a (local optimum. Practically, local neigbouroods tat only ave few similar neigbouroods may still contain noise after one iteration (unless is cosen large enoug; but tis would cause oversmooting in oter regions. In tis paper, we propose to keep track of te noise variance at every location in te image and to remove te remainder of te noise as a post-processing step using a local filter (see Section 4.. As discussed in Section 3., te NLMeans algoritm can be considered to be te first iteration of te Jacobian optimization algoritm for te cost function in equation (3. Tis would suggest tat applying te NLMeans algoritm iteratively would furter decrease te cost function. Te coice of te Leclerc robust function is somewat arbitrary. We will see tat furter improvements can be acieved by using te Bisquare robust function. Te existing NLMeans algoritms assume tat te image noise is wite (uncorrelated, wile in most practical denoising applications te noise is correlated. In tis case, computation of te similarity based on te Euclidean distance is ampered wic eventually leads to a poor denoising performance. By extending te reasoning from Section 3, it now becomes possible to devise a NLMeans filter for coloured (correlated noise. Tese modifications will be described more in detail in te remainder of tis Section. 4.. Local filtering of remaining noise In some circumstances (e.g. non-repetitive structures, one iteration of te NLMeans filter may not remove all of te noise. Teoretically, an infinite number of neigbouroods is required in order to completely suppress te noise, wic is not possible for finite image dimensions. However, it is possible to compute te noise variance of eac estimated vector, based on equations (5 and (7: σx,i = Var[ˆx i ] N = (λ i g (y i y j Var [y j ]+ j=,j i λ i Var [y i ] N = σw λ i j=,j i g (y i y j + (3 wit λ i given by equation (6. Clearly, te noise variance depends on te position in te image, wic means tat we are dealing wit non-stationary noise. In teory, any algoritm for non-stationary noise can be used, suc as te ones presented in [6,]. Because before te aggregation, te output of te NLMeans filter is a set of vectors, it is beneficial to use a vector-based denoising algoritm and to aggregate afterwards. In tis paper, we adopt a locally adaptive basis of Principle Components, as in [3]. Tis metod assumes tat te Gaussian noise is stationary, but an extension to non-stationary noise is straigtforward, as we will sow next. First, we define C y,i as te local covariance matrix of y i, estimated as follows : Ĉ y,i = wit ˆμ i = R + R + R (y i+n μ i (y i+n μ i T n= R R n= R y i+n To find te PCA basis, we apply te diagonalization: Ĉ y,i = U i Λ i U T i (4 Te local covariance matrix of x i, denoted as C x,i can be estimated as follows: Ĉ x,i = (Ĉy,i σwi + ( = U i Λi σw I + UT i (5 were ( + replaces possible negative eigenvalues by a small positive number, suc tat te resulting matrix is positive definite (due to estimation errors of Ĉy,i, it is possible tat te difference Ĉy,i σ wi as negative eigenvalues. Te linear MMSE estimator is given by: ˆx i = ˆμ i + Ĉx,i (Ĉx,i + σ x,ii (ˆxi ˆμ i = ˆμ i + U i A i U T i (ˆx i ˆμ i (6 wit A i = ( ( (Λi Λ i σw I Iσ + w + + σ x,i I a diagonal (Wiener filter matrix. Because a direct implementation of te above formulas as a ig computational cost (te diagonalization in equation (4 needs to Tis expressions given ere are for -d signals, but can be easily extended to images by using a double summation.

5 (a Original (b Noisy (0.6dB wit, following te Jacobi algoritm for determining te step size: λ (n i = + ( N j=,j i g ˆx (n i (7 ˆx (n j Implementation of tis estimate is analogous to te implementation of (7, except tat te estimates from te previous iterations are used as input. At first sigt, te reader may incorrectly ave te impression tat te iterative NLMeans filter increases te computational cost by te number of iterations. Instead, te iterative filter offers a number of advantages: (c NLMeans-W witout post-processing filter (9.8dB (d NLMeans-W wit post-processing filter (30.53dB Figure. Denoising example of Barbara: te effect of using te proposed post-processing filter (a Crop out of te original image (b Image wit wite Gaussian noise (σ =5. (c Te result of te NLMeans filter witout post-processing filter (d Te result of te NLMeans filter wit post-processing filter. PSNR values are between parenteses. be performed for every pixel in te image, we only estimate te covariance matrix Ĉy,i at subsampled positions i and assume tis covariance matrix is piecewise constant. Also, for neigbouroods wit a sufficient number of similar blocks, te variance σx,i will be very small after te NLMeans filtering stage. In tis case do not apply te local filtering step. An example for te Barbara image is given in Figure. It can be seen tat te NLMeans filter removes most of te noise, except in regions wit limited repetitivity. In tis regions, te post-processing filter works excellent in removing te remainder of te noise in Figure c, wile retaining te stripes. 4.. An iterative NLMeans filter An iterative NLMeans filter can be obtained by performing te optimization in equation (3 iteratively. Terefore, we coose te observed image as an initial estimate, i.e. ˆx (0 j = y j,j =,..., N. For te n-t iteration (n >0, we find: ˆx (n i = ˆx (n i N λ (n i j= g ( ˆx (n i ( ˆx (n j ˆx (n i ˆx (n j A small searc window (e.g. 3 3 can be used instead of te wole image, wic dramatically reduces te computation time. Wen applying more iterations, information from outside te searc window will also be used, resulting in a complete nonlocal denoising tecnique. Potentially a better end solution can be found, because te cost function is furter reduced. A limitation of te non-iterative NLMeans filter is tat te weigts are computed directly based on te observed noisy image. As a result, te weigts are very sensitive to te image noise. In [3], tis problem is avoided by applying a roug denoising preprocessing step before selecting similar neigbouroods. However, tis tecnique as te problem tat by roug denoising, some details may be lost, potentially resulting in an incorrect selection of dissimilar blocks. In te iterative NLMeans algoritm, every iteration reduces te average noise variance, resulting in better weigt estimates, tereby increasing te overall denoising performance Te coice of te robust loss function As said before, te NLMeans algoritm proposed in [0, ] can be interpreted as te first Jacobian iteration of a robust estimation metod, tat uses te Leclerc loss function. In our earlier work [6] we noted tat te exponential form of g(x still assigns positive weigts to dissimilar neigbouroods. Even toug tese weigts are very small, te estimated pixel intensities can be severely biased due to many small contributions. We terefore proposed a preclassification based on te first tree statistical moment to exclude dissimilar blocks [6]. An alternative is to cange te sape of te robust function. An overview of some relevant robust functions are given in Table. We will now look at te caracteristics of te robust weigting functions more in detail. Te weigting function associated wit te BLUE estimator derived in Section 3. and te Caucy weigting function ave a very slow decay (see Figure a. Tey assign larger weigts to dissimilar blocks tan te Leclerc robust function, wic will eventually lead to oversmooting.

6 Te Leclerc weigting function as a faster decay, but still assigns positive weigts to dissimilar blocks. Te Andrews weigting function imposes a ard tresold to compare neigbouroods (te weigt is 0 as soon as a given tresold is exceeded, wile te Tukey and Bisquare weigting functions rater use a soft tresold (Figure b. Experimentally we found tat applying a soft tresold often improves te visual quality, in analogy to wavelet tresolding. To furter improve upon te Tukey and Bisquare weigting functions, we also modified te Bisquare robust function in order to ave a steeper slope (see Table and Figure b. In Section 6 we will report te improvement in PSNR by using tis robust function NLMeans filter for correlated noise In te previous Sections, we assumed tat te Gaussian noise is uncorrelated (wite. Applying te NLMeans filter witout modifications to images corrupted wit correlated noise often yields a poor denoising performance (see Section 6. Fortunately, a robust estimator for correlated noise can be obtained by replacing te Euclidean distance y i y j by te Maalanobis distance (y i y j T C w (y i y j tat takes te noise covariance matrix C w into account (see e.g. [4], giving te following estimator: ˆx i =argmin x N j= ( ρ C / w (x y j wit ( / te square root of a positive definite matrix. Te weigt function becomes: { ( g Cw / (y i y j i j w(i, j = (8 i = j Clearly, te correlatedness of te noise only affects te weigt function and not te final averaging. However, te matrix multiplication in equation (8 is still computationally expensive, given te large number of weigts to be computed. Terefore, we apply a prewitening linear filter operation to te noisy image. If te noise Power Spectral Density is given by H(l, ten we compute te prewitened image Y prewit i,i=,..., N as: Ỹ prewit (l =Ỹ (l max(ɛ, H(l wit Ỹ prewit (l, Ỹ (l te discrete Fourier transform of respectively Y prewit i and Y i,andɛ a small positive number to ensure stability (for te results in Section 6, ɛ = 0 4. Next, tis prewitened image is used to compute te weigts based on te Euclidean distance (see also Figure 3. Y i Prewitening Y prewit i Weigt function w(i, j NLMeans Figure 3. Block diagram for te suppression of correlated noise 5. SPEEDING UP THE NLMEANS FILTER Because te algoritmic complexity of te brute force NLMeansfilteronimagesisO(N (K +, wit N te number of pixels in te image, an efficient implementation is desirable. Our previous analysis in [6] revealed tat te main part of te computation time is taken by te weigt computation. 5.. Exploiting weigt symmetry As in [6], we reduce te computation time by approximately a factor by exploiting te fact tat weigt functions are symmetrical (i.e. w(i, j =w(j, i. Terefore, we keep track of a weigt normalization matrix and a accumulated contribution matrix (bot of te same size as te input image, and at te beginning of te algoritm, initialized wit zeros. Wen processing a pixel i wit te contribution of a pixel j, we add te products w(i, jy i and w(i, jy j to te accumulated contribution matrix at te pixel positions j and i respectively. Te weigt normalization matrix is also accumulated at te same pixel positions wit w(i, j. As a result, we only need to compute weigts of neigbouroods for j > i. Finally, we normalize te accumulated contribution matrix via element-wise division by te weigt normalization matrix, in order to obtain te estimated image. Tese concepts can also be applied to te vector-based NLMeans filter: ere te products w(i, jb(ky i+k and w(i, jb(ky j+k are added to te accumulated contribution matrix at positions j + k and i + k respectively, for k = K,..., K. At te same positions, te products w(i, jb(k are added to weigt normalization matrix. 5.. Fast computation of te weigt functions based on te Euclidean distance For te robust functions in Table, w(i, j is a function of te Euclidean distance y i y j. Wen considering a constant position difference Δi (i.e. j = i +Δi, te function β(i = y i y i+δi can be practically implemented using a moving average filter applied to te squared difference between te signals ˆXi

7 Type g(r ρ(r { { / r r > +log r r > BLUE = / = r r r Caucy =/ (+ r / = log ( + r [ ] {( r Tukey = 3 r 6 ( r r = 0 r > 6 r > { sin(π r / { π r / r Andrews = = π ( cos (π r / r 0 r > π r > Leclerc =exp ( r = exp ( r ( 3 ( 3 ( r Bisquare = r r = 6 r + r 0 r > 0 r > 8 ( 9 ( 9 ( r Modified Bisquare = r r = 8 r + r 0 r > 0 r > Table. Overview of multivariate robust M-functions for extending te NLMeans algoritm, wit a smooting parameter Caucy Le Clerc Andrews "BLUE" Le Clerc Modified bisquare Tukey Bisquare (a (b Figure. Comparison of te weigting functions g(r for commonly used robust functions. Here, r = y i y j is te Euclidean distances between two vectors y i and y j

8 Y i and Y i+δi, requiring only pixel accesses instead of K +(in -d. For images, we use te straigtforward - d extension of te moving averaging filter. Togeter wit te weigt symmetry, tis brings te complexity down to approximately O(N, yielding a speed up of a factor (K + /. Wen considering square neigbouroods of size, K =5and te overall speedup is a factor / compared to our previous NLMeans filter in [6] Limited searc window Te limited searc window strategy assigns zero weigts to neigbouroods tat are too far away from eac oter, i.e. w(i, j =0if i j >N w. Tis tecnique may decrease te strengt of te NLMeans, especially for images wit many repetitive structures, but for many real images, te denoising performance is not significantly affected. Tis stems from te fact tat many similar blocks can be found in te local neigbourood. In tis paper, we use a limited searc window of 3 3 pixels, in order to keep te computational cost of te algoritm low. We remark again tat in combination wit te iterative approac from Section 4., te effective searc window is extended by N w in eac dimension, ence wen running at least N/ (N w iterations, te complete image will be searced Pseudo-code For illustrative purposes, te pseudo-code of te pixelbased -d implementation is sown in Figure 4. In te pseudo-code, te boundary andling is omitted for clarity, as well as te initialization (warm-up operations of te moving average filter. Te algoritm uses two loops : te first loop iterates on Δi in a limited searc window, te second loop runs over te wole signal. Te similarities corresponding to zero displacements (Δi =0are treated differently, because te weigts are easier to compute in tis case. It can be seen tat te tree acceleration tecniques proposed in tis Section, can be elegantly and efficiently combined in tis algoritm. We furter remark tat for te vector-based implementation (see Section, additional canges are required: a different weigt accumulation matrix needs to be used for eac position in te local neigbourood. 6. RESULTS AND DISCUSSION To assess te improvement gained by using te modified Bisquare cost function over te Leclerc cost function (see Table, we set up a denoising experiment wit 8 images and 9 noise levels. All images are corrupted wit artificially generated wite Gaussian noise wit varying noise levels, and subsequently te non-iterative NLMeans algoritm is applied to tem (including te post-processing filter. In Figure 5, we report te PSNR improvement by using te modified Bisquare cost function. We note tat on For images, four loops are needed (two outer loops and two inner loops for te x and y-directions. % outer loop (limited searc window weigt_cum = zeros(,n; weigt_norm = zeros(,n; for di=:nw % some initialization of te % moving average operation... % inner loop for j=:n i=j+di; % moving average (sum based computation % of te Euclidean distance eucl_dist=eucl_dist+(y(i+k-y(j+k^; eucl_dist=eucl_dist-(y(i-k--y(j-k-^; % weigt computation weigt=g(eucl_dist; % weigt accumulation weigt_cum(i = weigt_cum(i+weigt*y(j; weigt_norm(i = weigt_norm(i+weigt; % symmetry weigt_cum(j = weigt_cum(j+weigt*y(i; weigt_norm(j = weigt_norm(j+weigt; end; end; di=0; % zero displacement for i=:n weigt=; weigt_cum(i=weigt_cum(i+weigt*y(i; weigt_norm(i=weigt_norm(i+weigt; end; % final estimate x_at = weigt_cum./ weigt_norm; Figure 4. Pseudo-code for te improved NLMeans algoritm in -d (pixel-based implementation average, te improvement is approximately 0.dB. However te gain depends bot on te image as on te noise level, and is maximal for images wit many structures or strong edges, suc as Barbara and ouse and for large noise levels (e.g. σ 60. Because of te vast improvement for most images and/or noise levels, we will furter use te Bisquare cost function. Te results for te proposed NLMeans filter are generated using a neigbourood size of (i.e. K =5, a searc window of 3 3 (i.e. N w =5. Te Bisquare cost function is used and te -parameter of te robust function is selected experimentally as: =.σ. Te post-processing filter uses a subsampling factor 8 in te x and y direction for estimating te local covariance matrix (see Section 4.. To furter save computation time and to improve te numerical stability, te post-processing is done on neigbouroods of size 5 5, selected from te center of te neigbourood (ence te aggregation weigting function b(k = I( k /5 is used, wit I( te indicator function. In Table, te denoising performance of te proposed metod for wite noise (NLMeans-W is compared to BLS-GSM [5] (a state-ofte-art wavelet denoising tecnique and BM-3D (a stateof-te-art non-local denoising tecnique [3]. We give

9 Δ PSNR Average increment in PSNR Maximal increment in PSNR Minimal increment in PSNR σ Figure 5. Comparison of using te Leclerc cost function versus te modified Bisquare cost function, for a denoising experiment on a set of 8 images (Barbara, Lena, boats, couple, ill, man, peppers, ouse. Te values are te average (over te test set, maximal and minimal gain in PSNR by using te modified Bisquare cost function compared to te Leclerc cost function, for different noise levels σ. results for bot te non-iterative NLMeans (i.e. only one iteration and te iterative NLMeans filter as presented in Section 4., wit 5 iterations. Te iterative NLMeans filter brings an additional improvement compared to te non-iterative NLMeans. Tis is because te similarity weigts are refined iteratively, eventually resulting in a cleaner image wit more details. Te iterative NLMeans algoritm clearly outperforms te BLS-GSM filter and seems competitive to te BM-3D metod. Te visual performance of tese metods is compared in Figure 6-Figure 9. Here, te difference is astonising: te NLMeans filter reconstructs smooter images wit remarkably less artifacts tan te oter metods. Some details are better reconstructed, suc as te fine stripes in te at of Lena and te featers, even toug te PSNR of te NLMeans filter is sligtly lower compared to BM- 3D. Furter investigation revealed tat te inferior PSNR is caused by te reduced contrast (or oversmooting of fine structures, tat are usually visually ardly noticeable (for example, te tick stripes in te at of Lena located above te featers in Fig.6f. We also compared te denoising performance of te NLMeans extension to correlated noise (NLMeans-C to recent tecniques for correlated noise: BLS-GSM [5] and our recently proposed MP-GSM [5]. Te NLMeans filter brings a significant improvement bot visually as in PSNR: tere are almost no ringing artifacts and obviously no wavelet artifacts. We also compare to NLMeans-W (tat is not adapted to te correlated noise, just to illustrate tat te NLMeans filter, tat uses weigting functions based on te Euclidean distance, does not perform well in te presence of correlated noise. Te computation time for te non-iterative NLMeans filter is approximately 0s. on a Pentium IV processor and for a 5 5 grayscale image. Te iterative filter takes approximately 8 minutes, wen 5 iterations are used. Noise standard deviation LENA BLS-GSM [5] BM-3D [3] NLMeans-W (non-it NLMeans-W (5 it BARBARA BLS-GSM [5] BM-3D [3] NLMeans-W (non-it NLMeans-W (5 it Table. PSNR [db] results for wite noise 7. CONCLUSION By exploiting te repetitivity of structures in an image, a significant gain in denoising performance is obtained compared to purely local metods. We ave sown tat te NLMeans algoritm is te first iteration of te Jacobi optimization algoritm for robustly estimating te noisefree image, based on te Leclerc loss function. Next, we ave proposed several improvements to te NLMeans filter tat affect bot te visual quality and te computation time. Our proposed metod now compares favourably to state-of-te-art non-local denoising tecniques in PSNR and even offers a superior visual quality: te denoised images contain less artifacts witout sacrificing sarpness. Te proposed improvements can also be applied to oter image processing tasks suc as intra-frame superresolution, non-local demosaicing and deinterlacing. 8. REFERENCES [] L. Rudin and S. Oser, Total variation based image restoration wit free local constraints, in Proc. of IEEE International Conference on Image Processing (ICIP, vol., Nov. 994, pp [] C. Tomasi and R. Manduci, Bilateral filtering for gray and color images, in Proceedings International conference on computer vision, 998, pp [3] D. L. Donoo, De-Noising by Soft-Tresolding, IEEE Trans. Inform. Teory, vol. 4, pp , May 995. [4] L. Şendur and I. Selesnick, Bivariate srinkage wit local variance estimation, IEEE Signal Processing Letters, vol. 9, pp , 00. [5] J. Portilla, V. Strela, M. Wainwrigt, and E. Simoncelli, Image denoising using scale mixtures of gaussians in te wavelet domain, IEEE Transactions on image processing, vol., no., pp , 003. [6] J. Portilla, Image restoration using Gaussian Scale Mixtures in Overcomplete Oriented Pyramids (a review, in Wavelets XI, in SPIE s International Symposium on Optical Science and Tecnology, SPIE s 50t Annual Meeting, Proc. of te SPIE, vol. 594, San Diego, CA, aug 005, pp

10 [7] A. Pižurica and W. Pilips, Estimating te probability of te presence of a signal of interest in multiresolution single- and multiband image denoising, IEEE Transactions on image processing, vol. 5, no. 3, pp , 006. [8] F. Luisier, T. Blu, and M. Unser, A New SURE Approac to Image Denoising: Interscale Ortonormal Wavelet Tresolding, IEEE Trans. Image Processing, vol. 6, no. 3, pp , Mar [9] J. A. Guerrero-Colon, L. Mancera, and J. Portilla, Image restoration using space-variant gaussian scale mixtures in overcomplete pyramids, IEEE Trans. Image Proc., vol. 7, no., pp. 7 4, 008. [0] A. Buades, B. Coll, and J. Morel, A review of image denoising algoritms, wit a new one, SIAM interdisciplinary journal: Multiscale Modeling and Simulation, vol. 4, no., pp , 005. [] A. Buades, B. Coll., and J. Morel, A non local algoritm for image denoising, in Proc. Int. Conf. Computer Vision and Pattern Recognition (CVPR,vol., 005, pp [] M. Elad, On te Origin of te Bilateral Filter and Ways to Improve it, IEEE Transactions on Image Processing, vol., no. 0, pp. 4 5, 00. [3] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian, Image denoising by sparse 3d transform-domain collaborative filtering, IEEE Transactions on image processing, vol. 6, no. 8, pp , 007. [4] N. Azzabou, N. Paragias, and G. F., Image Denoising Based on Adapted Dictionary Computation, in Proc. of IEEE International Conference on Image Processing (ICIP, San Antonio, Texas, USA, Sept. 007, pp. 09. [9] M. Mamoudi and G. Sapiro, Fast image and video denoising via nonlocal means of similar neigboroods, IEEE Signal Processing Letters, vol., no., pp , Dec [0] J. Wang, Y. Guo, Y. Ying, Y. Liu, and Q. Peng, Fast non-local algoritm for image denoising, in Proc. of IEEE International Conference on Image Processing (ICIP, 006, pp [] B. R. C. and M. Vevilainen, Fast nonlocal means for image denoising, in Proc. SPIE Digital Potograpy III, R. A. Martin, J. M. DiCarlo, and N. Sampat, Eds., vol. 650, no.. SPIE, 007. [] B. Goossens, A. Pižurica, and W. Pilips, Wavelet domain image denoising for non-stationary and signal-dependent noise, in IEEE International Conference on Image Processing (ICIP, Atlanta, GA, USA, oct 006, pp [3] D. D. Muresan and T. W. Parks, Adaptive Principal Components and Image Denoising, in Proc. Int. Conf. on Image Processing (ICIP, 003. [4] N. A. Campbell, H. P. Lopuaä, and P. J. Rousseeuw, On te calculation of a robust s-estimator of a covariance matrix. Stat Med, vol. 7, no. 3, pp , Dec 998. [5] B. Goossens, A. Pižurica, and W. Pilips, Image Denoising Using Mixtures of Projected Gaussian Scale Mixtures, IEEE Transactions on Image Processing, 008, (submitted. [5] C. Kervrann, J. Boulanger, and P. Coupé, Bayesian Non-Local Means Filter, Image Redundancy and Adaptive Dictionaries for Noise Removal, in Proc. Int. Conf. on Scale Space and Variational Metods in Computer Visions (SSVM 07, Iscia, Italy, 007, pp [6] A. Dauwe, B. Goossens, H. Luong, and W. Pilips, A Fast Non-Local Image Denoising Algoritm, in Proc. SPIE Electronic Imaging, vol. 68, San José, USA, Jan 008. [7] C. Kervrann and J. Boulanger, Optimal spatial adaptation for patc-based image denoising, IEEE Trans. Image Processing, vol. 5, no. 0, pp , 006. [8] T. Brox and D. Cremers, Iterated Nonlocal Means for Texture Restoration, in Proc. Int. Conf. on Scale Space and Variational Metods in Computer Visions (SSVM 07, vol Iscia, Italy: Springer, LNCS, 007.

11 (a Original (b Noisy (0.9dB (c SV-GSM [9] (3.59dB (d MP-GSM [5] (3.8dB (e BM-3D [3] (3.06dB (f NLMeans-W (3.74dB Figure 6. Visual results for te Lena image corrupted wit wite noise (σ =5

12 (a Original (b Noisy (0.0dB (c BM-3D [3] (6.47dB (d NLMeans-W (6.59dB Figure 7. Visual results for te peppers image corrupted wit wite noise (σ =80

13 (a Original (b Noisy (8.95dB (a Original (b Noisy (8.95dB (c BLS-GSM [5] (9.33dB (d NLMeans-W (8.5dB (c BLS-GSM [5] (9.33dB (d NLMeans-W (8.5dB (e MP-GSM [5] (30.0dB (f NLMeans-C (30.74dB (e MP-GSM [5] (30.0dB (f NLMeans-C (30.74dB Figure 8. Visual results for correlated noise: two different crop-outs of te ouse image. PSNR values are between parenteses.

14 (a Original (b Noisy (6.09dB (a Original (b Noisy (6.09dB (c BLS-GSM (4.76dB (d NLMeans-W (.5dB (c BLS-GSM (4.76dB (d NLMeans-W (.5dB (e MP-GSM (5.0dB (f NLMeans-C (5.44dB (e MP-GSM (5.0dB (f NLMeans-C (5.44dB Figure 9. Visual results for correlated noise: two different crop-outs of te flinstones image. PSNR values are between parenteses.

Two Modifications of Weight Calculation of the Non-Local Means Denoising Method

Two Modifications of Weight Calculation of the Non-Local Means Denoising Method Engineering, 2013, 5, 522-526 ttp://dx.doi.org/10.4236/eng.2013.510b107 Publised Online October 2013 (ttp://www.scirp.org/journal/eng) Two Modifications of Weigt Calculation of te Non-Local Means Denoising

More information

A Fast Non-Local Image Denoising Algorithm

A Fast Non-Local Image Denoising Algorithm A Fast Non-Local Image Denoising Algorithm A. Dauwe, B. Goossens, H.Q. Luong and W. Philips IPI-TELIN-IBBT, Ghent University, Sint-Pietersnieuwstraat 41, 9000 Ghent, Belgium ABSTRACT In this paper we propose

More information

A GPU-accelerated real-time NLMeans algorithm for denoising color video sequences

A GPU-accelerated real-time NLMeans algorithm for denoising color video sequences A GPU-accelerated real-time NLMeans algorithm for denoising color video sequences Bart Goossens, Hiêp Luong, Jan Aelterman, Aleksandra Pižurica, and Wilfried Philips Ghent University, TELIN-IPI-IBBT St.-Pietersnieuwstraat

More information

Our Calibrated Model has No Predictive Value: An Example from the Petroleum Industry

Our Calibrated Model has No Predictive Value: An Example from the Petroleum Industry Our Calibrated Model as No Predictive Value: An Example from te Petroleum Industry J.N. Carter a, P.J. Ballester a, Z. Tavassoli a and P.R. King a a Department of Eart Sciences and Engineering, Imperial

More information

Vector Processing Contours

Vector Processing Contours Vector Processing Contours Andrey Kirsanov Department of Automation and Control Processes MAMI Moscow State Tecnical University Moscow, Russia AndKirsanov@yandex.ru A.Vavilin and K-H. Jo Department of

More information

Fast Calculation of Thermodynamic Properties of Water and Steam in Process Modelling using Spline Interpolation

Fast Calculation of Thermodynamic Properties of Water and Steam in Process Modelling using Spline Interpolation P R E P R N T CPWS XV Berlin, September 8, 008 Fast Calculation of Termodynamic Properties of Water and Steam in Process Modelling using Spline nterpolation Mattias Kunick a, Hans-Joacim Kretzscmar a,

More information

Structure-adaptive Image Denoising with 3D Collaborative Filtering

Structure-adaptive Image Denoising with 3D Collaborative Filtering , pp.42-47 http://dx.doi.org/10.14257/astl.2015.80.09 Structure-adaptive Image Denoising with 3D Collaborative Filtering Xuemei Wang 1, Dengyin Zhang 2, Min Zhu 2,3, Yingtian Ji 2, Jin Wang 4 1 College

More information

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING Idan Ram, Michael Elad and Israel Cohen Department of Electrical Engineering Department of Computer Science Technion - Israel Institute of Technology

More information

Novel speed up strategies for NLM Denoising With Patch Based Dictionaries

Novel speed up strategies for NLM Denoising With Patch Based Dictionaries Novel speed up strategies for NLM Denoising With Patch Based Dictionaries Rachita Shrivastava,Mrs Varsha Namdeo, Dr.Tripti Arjariya Abstract In this paper, a novel technique to speed-up a nonlocal means

More information

IMAGE RESTORATION VIA EFFICIENT GAUSSIAN MIXTURE MODEL LEARNING

IMAGE RESTORATION VIA EFFICIENT GAUSSIAN MIXTURE MODEL LEARNING IMAGE RESTORATION VIA EFFICIENT GAUSSIAN MIXTURE MODEL LEARNING Jianzhou Feng Li Song Xiaog Huo Xiaokang Yang Wenjun Zhang Shanghai Digital Media Processing Transmission Key Lab, Shanghai Jiaotong University

More information

Improved Non-Local Means Algorithm Based on Dimensionality Reduction

Improved Non-Local Means Algorithm Based on Dimensionality Reduction Improved Non-Local Means Algorithm Based on Dimensionality Reduction Golam M. Maruf and Mahmoud R. El-Sakka (&) Department of Computer Science, University of Western Ontario, London, Ontario, Canada {gmaruf,melsakka}@uwo.ca

More information

Unsupervised Learning for Hierarchical Clustering Using Statistical Information

Unsupervised Learning for Hierarchical Clustering Using Statistical Information Unsupervised Learning for Hierarcical Clustering Using Statistical Information Masaru Okamoto, Nan Bu, and Tosio Tsuji Department of Artificial Complex System Engineering Hirosima University Kagamiyama

More information

Image denoising in the wavelet domain using Improved Neigh-shrink

Image denoising in the wavelet domain using Improved Neigh-shrink Image denoising in the wavelet domain using Improved Neigh-shrink Rahim Kamran 1, Mehdi Nasri, Hossein Nezamabadi-pour 3, Saeid Saryazdi 4 1 Rahimkamran008@gmail.com nasri_me@yahoo.com 3 nezam@uk.ac.ir

More information

An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising

An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising Dr. B. R.VIKRAM M.E.,Ph.D.,MIEEE.,LMISTE, Principal of Vijay Rural Engineering College, NIZAMABAD ( Dt.) G. Chaitanya M.Tech,

More information

Patch-Based Color Image Denoising using efficient Pixel-Wise Weighting Techniques

Patch-Based Color Image Denoising using efficient Pixel-Wise Weighting Techniques Patch-Based Color Image Denoising using efficient Pixel-Wise Weighting Techniques Syed Gilani Pasha Assistant Professor, Dept. of ECE, School of Engineering, Central University of Karnataka, Gulbarga,

More information

Numerical Derivatives

Numerical Derivatives Lab 15 Numerical Derivatives Lab Objective: Understand and implement finite difference approximations of te derivative in single and multiple dimensions. Evaluate te accuracy of tese approximations. Ten

More information

Denoising an Image by Denoising its Components in a Moving Frame

Denoising an Image by Denoising its Components in a Moving Frame Denoising an Image by Denoising its Components in a Moving Frame Gabriela Ghimpețeanu 1, Thomas Batard 1, Marcelo Bertalmío 1, and Stacey Levine 2 1 Universitat Pompeu Fabra, Spain 2 Duquesne University,

More information

MAPI Computer Vision

MAPI Computer Vision MAPI Computer Vision Multiple View Geometry In tis module we intend to present several tecniques in te domain of te 3D vision Manuel Joao University of Mino Dep Industrial Electronics - Applications -

More information

A GEOMETRICAL WAVELET SHRINKAGE APPROACH FOR IMAGE DENOISING

A GEOMETRICAL WAVELET SHRINKAGE APPROACH FOR IMAGE DENOISING A GEOMETRICAL WAVELET SHRINKAGE APPROACH FOR IMAGE DENOISING Bruno Huysmans, Aleksandra Pižurica and Wilfried Philips TELIN Department, Ghent University Sint-Pietersnieuwstraat 4, 9, Ghent, Belgium phone:

More information

Implementation of Integral based Digital Curvature Estimators in DGtal

Implementation of Integral based Digital Curvature Estimators in DGtal Implementation of Integral based Digital Curvature Estimators in DGtal David Coeurjolly 1, Jacques-Olivier Lacaud 2, Jérémy Levallois 1,2 1 Université de Lyon, CNRS INSA-Lyon, LIRIS, UMR5205, F-69621,

More information

Haar Transform CS 430 Denbigh Starkey

Haar Transform CS 430 Denbigh Starkey Haar Transform CS Denbig Starkey. Background. Computing te transform. Restoring te original image from te transform 7. Producing te transform matrix 8 5. Using Haar for lossless compression 6. Using Haar

More information

4.2 The Derivative. f(x + h) f(x) lim

4.2 The Derivative. f(x + h) f(x) lim 4.2 Te Derivative Introduction In te previous section, it was sown tat if a function f as a nonvertical tangent line at a point (x, f(x)), ten its slope is given by te it f(x + ) f(x). (*) Tis is potentially

More information

Local features and image matching May 8 th, 2018

Local features and image matching May 8 th, 2018 Local features and image matcing May 8 t, 2018 Yong Jae Lee UC Davis Last time RANSAC for robust fitting Lines, translation Image mosaics Fitting a 2D transformation Homograpy 2 Today Mosaics recap: How

More information

Symmetric Tree Replication Protocol for Efficient Distributed Storage System*

Symmetric Tree Replication Protocol for Efficient Distributed Storage System* ymmetric Tree Replication Protocol for Efficient Distributed torage ystem* ung Cune Coi 1, Hee Yong Youn 1, and Joong up Coi 2 1 cool of Information and Communications Engineering ungkyunkwan University

More information

Mean Shifting Gradient Vector Flow: An Improved External Force Field for Active Surfaces in Widefield Microscopy.

Mean Shifting Gradient Vector Flow: An Improved External Force Field for Active Surfaces in Widefield Microscopy. Mean Sifting Gradient Vector Flow: An Improved External Force Field for Active Surfaces in Widefield Microscopy. Margret Keuper Cair of Pattern Recognition and Image Processing Computer Science Department

More information

Classification of Osteoporosis using Fractal Texture Features

Classification of Osteoporosis using Fractal Texture Features Classification of Osteoporosis using Fractal Texture Features V.Srikant, C.Dines Kumar and A.Tobin Department of Electronics and Communication Engineering Panimalar Engineering College Cennai, Tamil Nadu,

More information

BM3D Image Denoising with Shape-Adaptive Principal Component Analysis

BM3D Image Denoising with Shape-Adaptive Principal Component Analysis Note: A Matlab implementantion of BM3D-SAPCA is now available as part of the BM3D.zip package at http://www.cs.tut.fi/~foi/gcf-bm3d BM3D Image Denoising with Shape-Adaptive Principal Component Analysis

More information

Block Matching and 3D Filtering for Image Denoising

Block Matching and 3D Filtering for Image Denoising www.ijemr.net ISSN (ONLINE): 50-0758, ISSN (PRINT): 394-696 Volume-5, Issue-, February-05 International Journal of Engineering and Management Research Page Number: 308-33 Block Matching and Filtering for

More information

Image Restoration Using DNN

Image Restoration Using DNN Image Restoration Using DNN Hila Levi & Eran Amar Images were taken from: http://people.tuebingen.mpg.de/burger/neural_denoising/ Agenda Domain Expertise vs. End-to-End optimization Image Denoising and

More information

Image Interpolation using Collaborative Filtering

Image Interpolation using Collaborative Filtering Image Interpolation using Collaborative Filtering 1,2 Qiang Guo, 1,2,3 Caiming Zhang *1 School of Computer Science and Technology, Shandong Economic University, Jinan, 250014, China, qguo2010@gmail.com

More information

MATH 5a Spring 2018 READING ASSIGNMENTS FOR CHAPTER 2

MATH 5a Spring 2018 READING ASSIGNMENTS FOR CHAPTER 2 MATH 5a Spring 2018 READING ASSIGNMENTS FOR CHAPTER 2 Note: Tere will be a very sort online reading quiz (WebWork) on eac reading assignment due one our before class on its due date. Due dates can be found

More information

Density Estimation Over Data Stream

Density Estimation Over Data Stream Density Estimation Over Data Stream Aoying Zou Dept. of Computer Science, Fudan University 22 Handan Rd. Sangai, 2433, P.R. Cina ayzou@fudan.edu.cn Ziyuan Cai Dept. of Computer Science, Fudan University

More information

Image Restoration and Background Separation Using Sparse Representation Framework

Image Restoration and Background Separation Using Sparse Representation Framework Image Restoration and Background Separation Using Sparse Representation Framework Liu, Shikun Abstract In this paper, we introduce patch-based PCA denoising and k-svd dictionary learning method for the

More information

Bounding Tree Cover Number and Positive Semidefinite Zero Forcing Number

Bounding Tree Cover Number and Positive Semidefinite Zero Forcing Number Bounding Tree Cover Number and Positive Semidefinite Zero Forcing Number Sofia Burille Mentor: Micael Natanson September 15, 2014 Abstract Given a grap, G, wit a set of vertices, v, and edges, various

More information

Piecewise Polynomial Interpolation, cont d

Piecewise Polynomial Interpolation, cont d Jim Lambers MAT 460/560 Fall Semester 2009-0 Lecture 2 Notes Tese notes correspond to Section 4 in te text Piecewise Polynomial Interpolation, cont d Constructing Cubic Splines, cont d Having determined

More information

More on Functions and Their Graphs

More on Functions and Their Graphs More on Functions and Teir Graps Difference Quotient ( + ) ( ) f a f a is known as te difference quotient and is used exclusively wit functions. Te objective to keep in mind is to factor te appearing in

More information

Some Handwritten Signature Parameters in Biometric Recognition Process

Some Handwritten Signature Parameters in Biometric Recognition Process Some Handwritten Signature Parameters in Biometric Recognition Process Piotr Porwik Institute of Informatics, Silesian Uniersity, Bdziska 39, 41- Sosnowiec, Poland porwik@us.edu.pl Tomasz Para Institute

More information

IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS

IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS P.Mahalakshmi 1, J.Muthulakshmi 2, S.Kannadhasan 3 1,2 U.G Student, 3 Assistant Professor, Department of Electronics

More information

4.1 Tangent Lines. y 2 y 1 = y 2 y 1

4.1 Tangent Lines. y 2 y 1 = y 2 y 1 41 Tangent Lines Introduction Recall tat te slope of a line tells us ow fast te line rises or falls Given distinct points (x 1, y 1 ) and (x 2, y 2 ), te slope of te line troug tese two points is cange

More information

Image denoising with block-matching and 3D filtering

Image denoising with block-matching and 3D filtering Image denoising with block-matching and 3D filtering Kostadin Dabov, Alessandro Foi, Vladimir Katkovnik, and Karen Egiazarian Institute of Signal Processing, Tampere University of Technology, Finland PO

More information

Image Denoising using SWT 2D Wavelet Transform

Image Denoising using SWT 2D Wavelet Transform IJSTE - International Journal of Science Technology & Engineering Volume 3 Issue 01 July 2016 ISSN (online): 2349-784X Image Denoising using SWT 2D Wavelet Transform Deep Singh Bedi Department of Electronics

More information

2.8 The derivative as a function

2.8 The derivative as a function CHAPTER 2. LIMITS 56 2.8 Te derivative as a function Definition. Te derivative of f(x) istefunction f (x) defined as follows f f(x + ) f(x) (x). 0 Note: tis differs from te definition in section 2.7 in

More information

16th European Signal Processing Conference (EUSIPCO 2008), Lausanne, Switzerland, August 25-29, 2008, copyright by EURASIP

16th European Signal Processing Conference (EUSIPCO 2008), Lausanne, Switzerland, August 25-29, 2008, copyright by EURASIP 16t European Signal Processing Conference (EUSIPCO 008), Lausanne, Switzerland, August 5-9, 008, copyrigt by EURASIP ADAPTIVE WINDOW FOR LOCAL POLYNOMIAL REGRESSION FROM NOISY NONUNIFORM SAMPLES A. Sreenivasa

More information

Linear Interpolating Splines

Linear Interpolating Splines Jim Lambers MAT 772 Fall Semester 2010-11 Lecture 17 Notes Tese notes correspond to Sections 112, 11, and 114 in te text Linear Interpolating Splines We ave seen tat ig-degree polynomial interpolation

More information

Multi-Stack Boundary Labeling Problems

Multi-Stack Boundary Labeling Problems Multi-Stack Boundary Labeling Problems Micael A. Bekos 1, Micael Kaufmann 2, Katerina Potika 1 Antonios Symvonis 1 1 National Tecnical University of Atens, Scool of Applied Matematical & Pysical Sciences,

More information

Locally Adaptive Regression Kernels with (many) Applications

Locally Adaptive Regression Kernels with (many) Applications Locally Adaptive Regression Kernels with (many) Applications Peyman Milanfar EE Department University of California, Santa Cruz Joint work with Hiro Takeda, Hae Jong Seo, Xiang Zhu Outline Introduction/Motivation

More information

A Practical Approach of Selecting the Edge Detector Parameters to Achieve a Good Edge Map of the Gray Image

A Practical Approach of Selecting the Edge Detector Parameters to Achieve a Good Edge Map of the Gray Image Journal of Computer Science 5 (5): 355-362, 2009 ISSN 1549-3636 2009 Science Publications A Practical Approac of Selecting te Edge Detector Parameters to Acieve a Good Edge Map of te Gray Image 1 Akram

More information

RESTORING degraded images has been widely targeted

RESTORING degraded images has been widely targeted JOURNAL OF IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. XX, NO. X, 201X 1 Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering Milad Niknejad, Hossein Rabbani*, Senior

More information

Redundancy Awareness in SQL Queries

Redundancy Awareness in SQL Queries Redundancy Awareness in QL Queries Bin ao and Antonio Badia omputer Engineering and omputer cience Department University of Louisville bin.cao,abadia @louisville.edu Abstract In tis paper, we study QL

More information

Optimal In-Network Packet Aggregation Policy for Maximum Information Freshness

Optimal In-Network Packet Aggregation Policy for Maximum Information Freshness 1 Optimal In-etwork Packet Aggregation Policy for Maimum Information Fresness Alper Sinan Akyurek, Tajana Simunic Rosing Electrical and Computer Engineering, University of California, San Diego aakyurek@ucsd.edu,

More information

Cubic smoothing spline

Cubic smoothing spline Cubic smooting spline Menu: QCExpert Regression Cubic spline e module Cubic Spline is used to fit any functional regression curve troug data wit one independent variable x and one dependent random variable

More information

The Euler and trapezoidal stencils to solve d d x y x = f x, y x

The Euler and trapezoidal stencils to solve d d x y x = f x, y x restart; Te Euler and trapezoidal stencils to solve d d x y x = y x Te purpose of tis workseet is to derive te tree simplest numerical stencils to solve te first order d equation y x d x = y x, and study

More information

Investigating an automated method for the sensitivity analysis of functions

Investigating an automated method for the sensitivity analysis of functions Investigating an automated metod for te sensitivity analysis of functions Sibel EKER s.eker@student.tudelft.nl Jill SLINGER j..slinger@tudelft.nl Delft University of Tecnology 2628 BX, Delft, te Neterlands

More information

Traffic Sign Classification Using Ring Partitioned Method

Traffic Sign Classification Using Ring Partitioned Method Traffic Sign Classification Using Ring Partitioned Metod Aryuanto Soetedjo and Koici Yamada Laboratory for Management and Information Systems Science, Nagaoa University of Tecnology 603- Kamitomioamaci,

More information

ANTENNA SPHERICAL COORDINATE SYSTEMS AND THEIR APPLICATION IN COMBINING RESULTS FROM DIFFERENT ANTENNA ORIENTATIONS

ANTENNA SPHERICAL COORDINATE SYSTEMS AND THEIR APPLICATION IN COMBINING RESULTS FROM DIFFERENT ANTENNA ORIENTATIONS NTNN SPHRICL COORDINT SSTMS ND THIR PPLICTION IN COMBINING RSULTS FROM DIFFRNT NTNN ORINTTIONS llen C. Newell, Greg Hindman Nearfield Systems Incorporated 133. 223 rd St. Bldg. 524 Carson, C 9745 US BSTRCT

More information

CESILA: Communication Circle External Square Intersection-Based WSN Localization Algorithm

CESILA: Communication Circle External Square Intersection-Based WSN Localization Algorithm Sensors & Transducers 2013 by IFSA ttp://www.sensorsportal.com CESILA: Communication Circle External Square Intersection-Based WSN Localization Algoritm Sun Hongyu, Fang Ziyi, Qu Guannan College of Computer

More information

UUV DEPTH MEASUREMENT USING CAMERA IMAGES

UUV DEPTH MEASUREMENT USING CAMERA IMAGES ABCM Symposium Series in Mecatronics - Vol. 3 - pp.292-299 Copyrigt c 2008 by ABCM UUV DEPTH MEASUREMENT USING CAMERA IMAGES Rogerio Yugo Takimoto Graduate Scool of Engineering Yokoama National University

More information

Mean Waiting Time Analysis in Finite Storage Queues for Wireless Cellular Networks

Mean Waiting Time Analysis in Finite Storage Queues for Wireless Cellular Networks Mean Waiting Time Analysis in Finite Storage ueues for Wireless ellular Networks J. YLARINOS, S. LOUVROS, K. IOANNOU, A. IOANNOU 3 A.GARMIS 2 and S.KOTSOOULOS Wireless Telecommunication Laboratory, Department

More information

An Interactive X-Ray Image Segmentation Technique for Bone Extraction

An Interactive X-Ray Image Segmentation Technique for Bone Extraction An Interactive X-Ray Image Segmentation Tecnique for Bone Extraction Cristina Stolojescu-Crisan and Stefan Holban Politenica University of Timisoara V. Parvan 2, 300223 Timisoara, Romania {cristina.stolojescu@etc.upt.ro

More information

Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering

Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering JOURNAL OF IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. XX, NO. X, 201X 1 Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering Milad Niknejad, Hossein Rabbani*, Senior

More information

Coarticulation: An Approach for Generating Concurrent Plans in Markov Decision Processes

Coarticulation: An Approach for Generating Concurrent Plans in Markov Decision Processes Coarticulation: An Approac for Generating Concurrent Plans in Markov Decision Processes Kasayar Roanimanes kas@cs.umass.edu Sridar Maadevan maadeva@cs.umass.edu Department of Computer Science, University

More information

Multi-View Clustering with Constraint Propagation for Learning with an Incomplete Mapping Between Views

Multi-View Clustering with Constraint Propagation for Learning with an Incomplete Mapping Between Views Multi-View Clustering wit Constraint Propagation for Learning wit an Incomplete Mapping Between Views Eric Eaton Bryn Mawr College Computer Science Department Bryn Mawr, PA 19010 eeaton@brynmawr.edu Marie

More information

Grid Adaptation for Functional Outputs: Application to Two-Dimensional Inviscid Flows

Grid Adaptation for Functional Outputs: Application to Two-Dimensional Inviscid Flows Journal of Computational Pysics 176, 40 69 (2002) doi:10.1006/jcp.2001.6967, available online at ttp://www.idealibrary.com on Grid Adaptation for Functional Outputs: Application to Two-Dimensional Inviscid

More information

Image Denoising Based on Hybrid Fourier and Neighborhood Wavelet Coefficients Jun Cheng, Songli Lei

Image Denoising Based on Hybrid Fourier and Neighborhood Wavelet Coefficients Jun Cheng, Songli Lei Image Denoising Based on Hybrid Fourier and Neighborhood Wavelet Coefficients Jun Cheng, Songli Lei College of Physical and Information Science, Hunan Normal University, Changsha, China Hunan Art Professional

More information

Image Denoising based on Sparse Representation and Dual Dictionary

Image Denoising based on Sparse Representation and Dual Dictionary ISSN: 49 8958, Volume-4 Issue-4, April 015 Image Denoising based on Sparse Representation and Dual Dictionary Anees G Nat, Sreeram G, Sarafudeen K, Sreeraj M C Abstract earning-based image denoising aims

More information

Image denoising with patch based PCA: local versus global

Image denoising with patch based PCA: local versus global DELEDALLE, SALMON, DALALYAN: PATCH BASED PCA 1 Image denoising with patch based PCA: local versus global Charles-Alban Deledalle http://perso.telecom-paristech.fr/~deledall/ Joseph Salmon http://www.math.jussieu.fr/~salmon/

More information

REVERSIBLE DATA HIDING USING IMPROVED INTERPOLATION TECHNIQUE

REVERSIBLE DATA HIDING USING IMPROVED INTERPOLATION TECHNIQUE International Researc Journal of Engineering and Tecnology (IRJET) e-issn: 2395-0056 REVERSIBLE DATA HIDING USING IMPROVED INTERPOLATION TECHNIQUE Devendra Kumar 1, Dr. Krisna Raj 2 (Professor in ECE Department

More information

Hash-Based Indexes. Chapter 11. Comp 521 Files and Databases Spring

Hash-Based Indexes. Chapter 11. Comp 521 Files and Databases Spring Has-Based Indexes Capter 11 Comp 521 Files and Databases Spring 2010 1 Introduction As for any index, 3 alternatives for data entries k*: Data record wit key value k

More information

Multi-scale Statistical Image Models and Denoising

Multi-scale Statistical Image Models and Denoising Multi-scale Statistical Image Models and Denoising Eero P. Simoncelli Center for Neural Science, and Courant Institute of Mathematical Sciences New York University http://www.cns.nyu.edu/~eero Multi-scale

More information

13.5 DIRECTIONAL DERIVATIVES and the GRADIENT VECTOR

13.5 DIRECTIONAL DERIVATIVES and the GRADIENT VECTOR 13.5 Directional Derivatives and te Gradient Vector Contemporary Calculus 1 13.5 DIRECTIONAL DERIVATIVES and te GRADIENT VECTOR Directional Derivatives In Section 13.3 te partial derivatives f x and f

More information

Principal component analysis and Steerable Pyramid Transform Image Denoising Method for Gray Scale

Principal component analysis and Steerable Pyramid Transform Image Denoising Method for Gray Scale Principal component analysis and Steerable Pyramid Transform Image Denoising Method for Gray Scale Prachi Tiwari, Ayoush Johari, Dr. Soni Changlani Abstract This paper suggests a spatially adaptive image

More information

Edge Patch Based Image Denoising Using Modified NLM Approach

Edge Patch Based Image Denoising Using Modified NLM Approach Edge Patch Based Image Denoising Using Modified NLM Approach Rahul Kumar Dongardive 1, Ritu Shukla 2, Dr. Y K Jain 3 1 Research Scholar of SATI Vidisha, M.P., India 2 Asst. Prof., Computer Science and

More information

MULTIVIEW 3D VIDEO DENOISING IN SLIDING 3D DCT DOMAIN

MULTIVIEW 3D VIDEO DENOISING IN SLIDING 3D DCT DOMAIN 20th European Signal Processing Conference (EUSIPCO 2012) Bucharest, Romania, August 27-31, 2012 MULTIVIEW 3D VIDEO DENOISING IN SLIDING 3D DCT DOMAIN 1 Michal Joachimiak, 2 Dmytro Rusanovskyy 1 Dept.

More information

SUPPLEMENTARY MATERIAL

SUPPLEMENTARY MATERIAL SUPPLEMENTARY MATERIAL Zhiyuan Zha 1,3, Xin Liu 2, Ziheng Zhou 2, Xiaohua Huang 2, Jingang Shi 2, Zhenhong Shang 3, Lan Tang 1, Yechao Bai 1, Qiong Wang 1, Xinggan Zhang 1 1 School of Electronic Science

More information

Section 1.2 The Slope of a Tangent

Section 1.2 The Slope of a Tangent Section 1.2 Te Slope of a Tangent You are familiar wit te concept of a tangent to a curve. Wat geometric interpretation can be given to a tangent to te grap of a function at a point? A tangent is te straigt

More information

Image Denoising using Locally Learned Dictionaries

Image Denoising using Locally Learned Dictionaries Image Denoising using Locally Learned Dictionaries Priyam Chatterjee and Peyman Milanfar Department of Electrical Engineering, University of California, Santa Cruz, CA 95064, USA. ABSTRACT In this paper

More information

An Algorithm for Loopless Deflection in Photonic Packet-Switched Networks

An Algorithm for Loopless Deflection in Photonic Packet-Switched Networks An Algoritm for Loopless Deflection in Potonic Packet-Switced Networks Jason P. Jue Center for Advanced Telecommunications Systems and Services Te University of Texas at Dallas Ricardson, TX 75083-0688

More information

Australian Journal of Basic and Applied Sciences. Partial Differential Equation Based Denoising Technique for MRI Brain Images

Australian Journal of Basic and Applied Sciences. Partial Differential Equation Based Denoising Technique for MRI Brain Images AENSI Journals Australian Journal of Basic and Applied Sciences Journal ome page: www.ajbasweb.com Partial Differential Equation Based Denoising Tecnique for MRI Brain Images 1 S. Jansi and P. Subasini

More information

A VOLUME-BASED METHOD FOR DENOISING ON CURVED SURFACES

A VOLUME-BASED METHOD FOR DENOISING ON CURVED SURFACES A VOLUME-BASED METHOD FOR DENOISING ON CURVED SURFACES Harry Biddle Double Negative Visual Effects London, UK b@dneg.com Ingrid von Glen, Colin B. Macdonald, Tomas März Matematical Institute University

More information

VIDEO DENOISING BASED ON ADAPTIVE TEMPORAL AVERAGING

VIDEO DENOISING BASED ON ADAPTIVE TEMPORAL AVERAGING Engineering Review Vol. 32, Issue 2, 64-69, 2012. 64 VIDEO DENOISING BASED ON ADAPTIVE TEMPORAL AVERAGING David BARTOVČAK Miroslav VRANKIĆ Abstract: This paper proposes a video denoising algorithm based

More information

RECONSTRUCTING OF A GIVEN PIXEL S THREE- DIMENSIONAL COORDINATES GIVEN BY A PERSPECTIVE DIGITAL AERIAL PHOTOS BY APPLYING DIGITAL TERRAIN MODEL

RECONSTRUCTING OF A GIVEN PIXEL S THREE- DIMENSIONAL COORDINATES GIVEN BY A PERSPECTIVE DIGITAL AERIAL PHOTOS BY APPLYING DIGITAL TERRAIN MODEL IV. Évfolyam 3. szám - 2009. szeptember Horvát Zoltán orvat.zoltan@zmne.u REONSTRUTING OF GIVEN PIXEL S THREE- DIMENSIONL OORDINTES GIVEN Y PERSPETIVE DIGITL ERIL PHOTOS Y PPLYING DIGITL TERRIN MODEL bsztrakt/bstract

More information

Locally Adaptive Learning for Translation-Variant MRF Image Priors

Locally Adaptive Learning for Translation-Variant MRF Image Priors Locally Adaptive Learning for Translation-Variant MRF Image Priors Masayuki Tanaka and Masatoshi Okutomi Tokyo Institute of Technology 2-12-1 O-okayama, Meguro-ku, Tokyo, JAPAN mtanaka@ok.ctrl.titech.ac.p,

More information

Expected Patch Log Likelihood with a Sparse Prior

Expected Patch Log Likelihood with a Sparse Prior Expected Patch Log Likelihood with a Sparse Prior Jeremias Sulam and Michael Elad Computer Science Department, Technion, Israel {jsulam,elad}@cs.technion.ac.il Abstract. Image priors are of great importance

More information

Proceedings of the 8th WSEAS International Conference on Neural Networks, Vancouver, British Columbia, Canada, June 19-21,

Proceedings of the 8th WSEAS International Conference on Neural Networks, Vancouver, British Columbia, Canada, June 19-21, Proceedings of te 8t WSEAS International Conference on Neural Networks, Vancouver, Britis Columbia, Canada, June 9-2, 2007 3 Neural Network Structures wit Constant Weigts to Implement Dis-Jointly Removed

More information

PYRAMID FILTERS BASED ON BILINEAR INTERPOLATION

PYRAMID FILTERS BASED ON BILINEAR INTERPOLATION PYRAMID FILTERS BASED ON BILINEAR INTERPOLATION Martin Kraus Computer Grapics and Visualization Group, Tecnisce Universität Müncen, Germany krausma@in.tum.de Magnus Strengert Visualization and Interactive

More information

Section 2.3: Calculating Limits using the Limit Laws

Section 2.3: Calculating Limits using the Limit Laws Section 2.3: Calculating Limits using te Limit Laws In previous sections, we used graps and numerics to approimate te value of a it if it eists. Te problem wit tis owever is tat it does not always give

More information

Total Variation Denoising with Overlapping Group Sparsity

Total Variation Denoising with Overlapping Group Sparsity 1 Total Variation Denoising with Overlapping Group Sparsity Ivan W. Selesnick and Po-Yu Chen Polytechnic Institute of New York University Brooklyn, New York selesi@poly.edu 2 Abstract This paper describes

More information

Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling

Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling Moritz Baecher May 15, 29 1 Introduction Edge-preserving smoothing and super-resolution are classic and important

More information

Utilizing Call Admission Control to Derive Optimal Pricing of Multiple Service Classes in Wireless Cellular Networks

Utilizing Call Admission Control to Derive Optimal Pricing of Multiple Service Classes in Wireless Cellular Networks Utilizing Call Admission Control to Derive Optimal Pricing of Multiple Service Classes in Wireless Cellular Networks Okan Yilmaz and Ing-Ray Cen Computer Science Department Virginia Tec {oyilmaz, ircen}@vt.edu

More information

Advanced phase retrieval: maximum likelihood technique with sparse regularization of phase and amplitude

Advanced phase retrieval: maximum likelihood technique with sparse regularization of phase and amplitude Advanced phase retrieval: maximum likelihood technique with sparse regularization of phase and amplitude A. Migukin *, V. atkovnik and J. Astola Department of Signal Processing, Tampere University of Technology,

More information

Image Registration via Particle Movement

Image Registration via Particle Movement Image Registration via Particle Movement Zao Yi and Justin Wan Abstract Toug fluid model offers a good approac to nonrigid registration wit large deformations, it suffers from te blurring artifacts introduced

More information

Alternating Direction Implicit Methods for FDTD Using the Dey-Mittra Embedded Boundary Method

Alternating Direction Implicit Methods for FDTD Using the Dey-Mittra Embedded Boundary Method Te Open Plasma Pysics Journal, 2010, 3, 29-35 29 Open Access Alternating Direction Implicit Metods for FDTD Using te Dey-Mittra Embedded Boundary Metod T.M. Austin *, J.R. Cary, D.N. Smite C. Nieter Tec-X

More information

3.6 Directional Derivatives and the Gradient Vector

3.6 Directional Derivatives and the Gradient Vector 288 CHAPTER 3. FUNCTIONS OF SEVERAL VARIABLES 3.6 Directional Derivatives and te Gradient Vector 3.6.1 Functions of two Variables Directional Derivatives Let us first quickly review, one more time, te

More information

MAP MOSAICKING WITH DISSIMILAR PROJECTIONS, SPATIAL RESOLUTIONS, DATA TYPES AND NUMBER OF BANDS 1. INTRODUCTION

MAP MOSAICKING WITH DISSIMILAR PROJECTIONS, SPATIAL RESOLUTIONS, DATA TYPES AND NUMBER OF BANDS 1. INTRODUCTION MP MOSICKING WITH DISSIMILR PROJECTIONS, SPTIL RESOLUTIONS, DT TYPES ND NUMBER OF BNDS Tyler J. lumbaug and Peter Bajcsy National Center for Supercomputing pplications 605 East Springfield venue, Campaign,

More information

Image denoising using curvelet transform: an approach for edge preservation

Image denoising using curvelet transform: an approach for edge preservation Journal of Scientific & Industrial Research Vol. 3469, January 00, pp. 34-38 J SCI IN RES VOL 69 JANUARY 00 Image denoising using curvelet transform: an approach for edge preservation Anil A Patil * and

More information

Image Modeling and Denoising With Orientation-Adapted Gaussian Scale Mixtures David K. Hammond and Eero P. Simoncelli, Senior Member, IEEE

Image Modeling and Denoising With Orientation-Adapted Gaussian Scale Mixtures David K. Hammond and Eero P. Simoncelli, Senior Member, IEEE IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 17, NO. 11, NOVEMBER 2008 2089 Image Modeling and Denoising With Orientation-Adapted Gaussian Scale Mixtures David K. Hammond and Eero P. Simoncelli, Senior

More information

Curvelet Transform with Adaptive Tiling

Curvelet Transform with Adaptive Tiling Curvelet Transform with Adaptive Tiling Hasan Al-Marzouqi and Ghassan AlRegib School of Electrical and Computer Engineering Georgia Institute of Technology, Atlanta, GA, 30332-0250 {almarzouqi, alregib}@gatech.edu

More information

A New Orthogonalization of Locality Preserving Projection and Applications

A New Orthogonalization of Locality Preserving Projection and Applications A New Orthogonalization of Locality Preserving Projection and Applications Gitam Shikkenawis 1,, Suman K. Mitra, and Ajit Rajwade 2 1 Dhirubhai Ambani Institute of Information and Communication Technology,

More information

An Overview of New Features in

An Overview of New Features in 6. LS-DYNA Anwenderforum, Frankental 2007 Optimierung An Overview of New Features in LS-OPT Version 3.3 Nielen Stander*, Tusar Goel*, David Björkevik** *Livermore Software Tecnology Corporation, Livermore,

More information

Centralized Sparse Representation for Image Restoration

Centralized Sparse Representation for Image Restoration Centralized Sparse Representation for Image Restoration Weisheng Dong Sch. of Elec. Engineering Xidian University, China wsdong@mail.xidian.edu.cn Lei Zhang Dept. of Computing The Hong Kong Polytechnic

More information