Image contrast enhancement based on local brightness and contouring artifact improvement for large-scale LCD TVs JONG-HEE HWANG 1,2, JEAN Y. SONG 1, YOON-SIK CHOE 1 1 Department of Electrical and Electronics Engineering, Yonsei University, Seoul, REPUBLIC of KOREA 2 Preceding Development Division, LG Display Co., LTD, Paju, REPUBLIC of KOREA jonghee38@yonsei.ac.r, jeanyoung@yonsei.ac.r, yschoe@yonsei.ac.r Abstract: - Generally, the global histogram equalization of contrast enhancement methods was used in various application fields because of its simple and effective function. But, as it has a disadvantage that the brightness of an image is changed excessively, in spite of heavy computational complexity, overlapped sub-bloc histogram equalization method reflecting local histogram properties is used mainly. Also, the global and local histogram equalization methods increase excessively the contrast of simple bacground occupying wide scope, and they causes the false contour which is unpleasant to the perception operation of an observer. So, in order to improve the local brightness, this paper derived the local histogram equalization (LHE) with the minimal overlapped bloc size that is optimized for Full High Definition (FHD: 1920 1080) representing the standard image size of large-size LCD (Liquid Crystal Display) TV and can be applied without causing the blocing artifacts, and then it proposed the image contrast enhancement method that combines the modified error diffusion (ED) method in order to reduce the contouring artifacts. This experimental results show that the proposed contrast enhancement method preserves the local image brightness and suppresses the contouring artifacts. Key-Words: - Histogram equalization, Local brightness, Contouring artifacts, False contour, Sub-bloc overlap. 1 Introduction Contrast enhancement is an image process technique of a low step that it does clearly an area of interest at images or redistributes intensity values to improve image quality. So, it clarifies the visual difference between dar and bright areas of an image. If the contrast of an image is increased, an observer can see a pictorial image in more detail. This is the pure perception operation that the gross amount of information never increases at images. Our perception operation is more sensitive to the contrast of brightness rather than the intensity of a pure brightness. The most popular of these methods is called histogram equalization. Histogram equalization reassigns the brightness values of pixels based on the image histogram. Individual pixels retain their brightness order (that is, they remain brighter or darer than other pixels) but the values are shifted, so that an equal number of pixels have each possible brightness value. In many cases, this spreads out the values in regions where different regions meet, showing detail in areas with a high brightness gradient. An image having a few regions with very similar brightness values presents a histogram with peas. The sizes of these peas give the relative area of the different phase regions and are useful for image analysis [1]. Histogram equalization techniques can be classified greatly in ways to use global information and local information for input images. As global histogram equalization doesn't tae space information of each part at images into consideration and uses histogram information of the entire image, it is hard to improve a local contrast value. In addition, as it converts the average brightness of the image to the middle brightness value of the image by redistributing the brightness value of the image, it causes the phenomenon that bright areas of the image becomes hazy [2 3]. Local histogram equalization (LHE) technique has been used primarily to overcome these problems with global contrast enhancement. A number of ways including AHE (Adaptive Histogram Equalizatio [4], POSHE (Partially Overlapped Sub-bloc Histogram Equalizatio [5], BBHE (Brightness preserving Bi-Histogram Equalizatio [6], RSIHE (Recursive Sub-Image Histogram Equalizatio [7] have been proposed. However, the various methods mentioned above must decide on the size of a partition bloc and the size of an overlapped step in common and need a lot of process time in proportion to the resolution of the image and the density of histogram. Also, as the variation of pixel values increases in the flat area that the brightness should change gradually, the false contour is generated. So, in order to improve the local brightness, this paper derived the LHE with the minimal overlapped bloc size that is optimized for FHD representing the standard image size of large-size LCD TV and can be applied without causing the blocing artifacts, and then it ISBN: 978-960-474-262-2 154
1 1 Recent Researches in Circuits, Systems, Electronics, Control & Signal Processing proposed the image contrast enhancement method that combines the modified error diffusion (ED) method in order to improve the contouring artifacts. The rest of this paper is organized as follows. Section 2 explains the previous research for contrast enhancement algorithm. Section 3 presents the error diffusion technique. Our proposed contrast enhancement method is present in Section 4. Section 5 describes the experimental results. Finally, we conclude in Section 6. 2 Histogram Equalization Overview This section summarizes fundamental bacground of global histogram equalization (GHE) and bloc overlapped histogram equalization typically being used in local histogram equalization (LHE). 2.1 Global Histogram Equalization (GHE) If a given input image is defined by X={X(i,j) X(i,j) {X 0,X 1,,X L-1 }, it can be consists of L brightness levels. X (i, j) represents the brightness of a given image in the spatial position. The probability density function for X is defined as equation (1). p X ( X 0 X ) = 1, n n L = 1 0 p X ( X ) = 1 (1) Here, is the range from 0 to L-1, n is the total number of pixels of a input image, n represents the number of occurrences of brightness value X. Next, the cumulative distribution function for brightness value X lie equation (2) can be obtained from the probability density function c ( x ) = p X ( X ) (2) j = 0 j Here, is the range from 1 to L-1, c(x L-1 ) is 1. As histogram equalization is a scheme that maps an input image into the entire dynamic area (X 0, X L-1 ) using equation (2), it can be expressed as the conversion function f (x) of equation (3). f x) = X + ( X L X ) c( ) (3) ( 0 1 0 x Using GHE in order to enhance the global contrast, GHE doesn t reflect brightness characteristics of local areas as shown in Fig. 1. In addition, as it converts the average brightness of the image to the middle brightness value of the image by redistributing the brightness value of the image, it causes the phenomenon that bright areas of the image becomes hazy. Especially, including a lot of areas with similar brightness, the contrast can be reduced or image quality can be damaged because we can t adjust locally the contrast. These problems are solved by using local contrast enhancement technique. 40000 35000 30000 25000 20000 15000 10000 5000 0 12 23 34 45 56 67 78 89 100 111 122 133 144 155 166 177 188 199 210 221 232 243 254 40000 35000 30000 25000 20000 15000 10000 5000 0 12 23 34 45 56 67 78 89 100 111 122 133 144 155 166 177 188 199 210 221 232 243 254 Fig.1 Contrast enhancement using global histogram equalization: (a) original image; (b) test results. 2.2 Bloc Overlapped Histogram Equalization The bloc overlapped histogram equalization is performed with the following procedures. First, a sub-bloc of appropriate size is set up. Second, the local histogram regarding the sub-bloc is obtained. Third, histogram equalization is performed by using this information in the central point of a relevant sub-bloc, and then the sub-bloc is moved by one pixel. In this way, as each pixel of an input image is redistributed by the local histogram of the neighboring sub-bloc, each point can be adapted in a brightness situation near this point and this method can increase the contrast of all objects and bacground. However, this method must perform the sub-bloc histogram equalization about all pixels of an input image. So, the number of iterations increases excessively. In order to reduce the repetition frequency of bloc overlapped histogram equalization, the number of overlap should be reduced. For this purpose, non-overlapped sub-bloc histogram equalization technique can be used. In this method, each sub-bloc is not overlapped with adjacent sub-blocs and it is histogram equalized by moving the origin of sub-bloc. However, this non-overlapped method causes the blocing effect. The reason for the blocing effect can be explained as follows. Considering adjacent sub-blocs in the input image, average brightness of adjacent sub-blocs has similar and the brightness of pixels at the borders of the sub-blocs changes gradually. So, the blocing artifacts don't appear. However, since brightness and size of objects are different within each sub-bloc, they have different types of histograms. At the result that non-overlapped sub-bloc histogram equalization is performed, the blocing artifacts are appeared on an adjacent border of the sub-blocs because ISBN: 978-960-474-262-2 155
two different sub-bloc histograms are used as transformation functions. Methods for reducing the number of iterations for sub-bloc overlapped histogram equalization have been proposed in reference papers [5],[7]. In other words, the histogram equalization function of an area is obtained by using the weighted sum for histograms of adjacent areas rather than only using the local histogram of relevant area. For example, if 3 3 mas is used, the histogram equalization function in the center of the area can be obtained from the histogram of its own and histograms of adjacent eight regions. Since this mas has the similar form with the low-pass filter for the image and its function is also similar, it blurs an image by reducing the brightness difference between adjacent pixels. Liewise, the form of a histogram equalization function will also be similar in each sub-bloc. As this result, these papers [5],[7] showed that the blocing artifacts are reduced experimentally. However, as the blocing artifacts can still occur depending on the image characteristics, these papers represented that post-processing such as blocing effect reduction filter must be conducted. Therefore, since the image size of this paper was determined as FHD resolution to enhance the contrast for large-size LCD TV, the pre-test is needed to derive parameters for the minimum bloc size which can be applied without causing the blocing artifacts. Through this experiment, we will be able to reduce the computational complexity of sub-bloc overlapped method and improve the blocing artifacts. 3 Error Diffusion Technique Although original image maintains high quality and good quality can be implemented due to high bit-depth, a case that has to reduce bit-depth of an input image can exist because of the expression ability restriction of display devices. In this case, the original image is quantized, the quantization error will occur. the first time by Floyd and Steinberg diffuses a quantization error caused by the binarization of pixel values to a surrounding pixel. The quantization error is multiplied by appropriate weighting value of ED mas. General ED algorithm can be represented by the following equations [8]. e( = X ED _ out ( u( (4) u( = X ( w( e( m, n l (5) X ED _ out ED_ in ) (, l) R 255, ( = Q( u( ) = 0, if u( T otherwise (6) Here, e( is the quantization error, u( is the state variable. X ED_out ( means the pixel values of binarized image and w( represents the weighting values of ED mas. Fig. 3 shows the ED mas of Floyd and Steinberg. The weighting values for each quantization error which are located on the right side of Fig.3 are modified in the ED algorithm of Floyd and Steinberg. When implemented in hardware, the product of coefficients can be implemented as simple shift operations without using multipliers. So, hardware designers will be able to significantly reduce the computational complexity of conventional ED by substituting shift operation for multiplication operation. Fig.3 Floyd-Steinberg s error diffusion mas. Fig. 4 shows the propagation flow for ED. Quantization error is propagated to the right pixel and the three pixels of next horizontal line. Here, looing at the four sides on first frame, the top left side is propagated in the three directions and the top right and bottom right sides are propagated in the two directions and one direction respectively. Conversely, a pixel receives the four error values from neighboring pixels. Fig.2 The bloc diagram of conventional error diffusion algorithm. In general, when the bit-depth that can display an image to have passed through a quantization process is insufficient, the way to be used to improve effectively image quality is error diffusion (ED) technique. As shown in Fig. 2, the ED algorithm that was proposed for Fig.4 Error diffusion propagation flow on first frame. ISBN: 978-960-474-262-2 156
In terms of product applications, ED technique has been used to improve the false contour of the PDP (Plasma Display Panel) due to the reduction of gray levels which can be expressed in dar areas [9]. 4 The Proposed Contrast Enhancement In order to improve the local brightness and the contouring artifacts together, this paper proposed the image contrast enhancement that combines the optimized histogram equalization and the modified ED methods. The whole architecture of proposed contrast enhancement as shown in Fig. 5 is composed of histogram equalization unit for the local brightness improvement, input parameter definition unit, and modified error diffusion unit. The decision base of sub-bloc size can be explained as follows. In LCD TV adopting LED (Light Emitting Diode) baclight, local dimming technique has been used in order to minimize power consumption and light leaage [10]. As shown in Fig. 6, after individual LED devices are arranged on the rear of the LCD TV and are divided into many blocs, the divided blocs are driven by local dimming technique according to the image characteristics. Here, PWM (Pulse Width Modulatio signal of LED devices belonging to each bloc is determined by using the histogram information of each bloc. So, when performing histogram equalization, we can use hardware resources effectively and get better contrast enhancement by using sub-bloc size exactly the same as local dimming bloc size. Fig.6 The driving system of LCD TV using LED baclight. Fig.5 The whole architecture of proposed contrast enhancement. First, histogram equalization unit consists of sub-bloc and step size decision bloc, histogram equalization through minimal sub-bloc overlap method, and I2C (Inter Integrated Circuit) communication unit that plays a role to control easily the main parameters through PC users from outside. Here, X in ( represents the input image with low contrast. In case of color images, it means the luminance signal that is obtained through color space conversion such as RGB to YCbCr or RGB to YCoCg. X HE_out ( represents the result of histogram equalization optimized for large size LCD TV, SS hv is sub-bloc size, S h and S v are horizontal and vertical step size respectively. Secondly, conventional ED algorithm was properly modified to reduce the contouring artifacts occurring after histogram equalization. The modified ED algorithm contains input parameter definition and modified error diffusion units as shown in Figure 5. In order to define the inputs which are used to the modified ED algorith average image (Y ) and scaling of difference image (Y DI ) is introduced newly. Modified ED process of Fig. 5 can be expressed as follows. e( = Y _ ( Y ' ( (7) ED out Y' ( = Y ( w( e( m, n l) (8) (, l) R ' ( + YDI u( = Y ( (9) YED _ out ( = Q( u( ) (10) Here, e( is the difference value that error image is spread by error diffusion, Y' ( is the state variable, and u( represents the image that will be quantized by the multi-level quantizer. Unlie conventional ED algorith intermediate variable is set as Y' (. ISBN: 978-960-474-262-2 157
After adding the average image including the result that the error image is diffused to the scaling of difference image, it is entered into the multi-level quantizer. The average gray level for input and output values isn t maintained because the modified ED disperses to neighboring pixels properly by multiplying the value of equation (7) and the weighting value of ED mas. Thus, the distortion such as false contour is reduced in flat area due to the reduction of difference values between neighboring pixels. 5 Experimental Results We implement the local histogram equalization (LHE) method having the optimized sub-bloc size and step size for large-size LCD TV, and then implement the proposed method that combines ED and LHE by using the visual C++ 6.0 environment. Simulations are conducted for FHD images with low contrast and various characteristics where the bit depth is 8-bit. As shown in Fig. 7, sub-bloc size of LHE (SS hv ) is 120 90 and it is equal to local dimming bloc size. Through repeated experiments, overlapped horizontal and vertical step sizes (S h, S v ) to remove the blocing artifacts were determined as 6 and 9 pixels respectively. We compare the performance from two inds of perspectives including visual quality and hardware complexity. 5.1 Visual Quality Comparison Fig.7 The sub-bloc for local histogram equalization. Fig. 8 shows the visual quality comparison for FHD image with low contrast. Fig. 8(b) is the enlarged image of the flat area in Fig. 8(a), Fig.8(C) is the enlarged result image performing LHE, and Fig. 8(d) represents the result image applying the modified error diffusion algorithm to Fig. 8(c). It is noticed that the contrast is improved through histogram information of the top right section in Fig. 8(c) and the blocing artifacts in the sub-bloc overlapped LHE isn t visible. However, the false contour problem occurred due to sharp change of pixel values in the flat area that the brightness value has to change gradually. As shown in Fig. 8(d), since the difference between the input image with low contrast and the image with improved contrast using LHE is dispersed to the neighboring pixels by ED technique, the pixel values are adjusted to change gradually in the flat area. So, we can see that the false contour is reduced. Fig.8 Visual quality comparison for FHD image with low contrast: (a) original image; (b) original image enlargement for a square area; (c) LHE results; (d) proposed method (the combination LHE and ED). ISBN: 978-960-474-262-2 158
5.2 Complexity Comparison The proposed method shows much lower complexity than bloc overlapped histogram equalization [11] using the overlap of pixel unit because it uses a wide range of step size. Next, when it is compared with POSHE [5] using the weighted sum for histograms of adjacent areas, a computational complexity of histogram equalization is the same. But, as POSHE uses the blocing effect reduction filter to prevent the blocing artifacts additionally, 3 line memory is needed. On the other hand, as the proposed method uses the modified ED technique to improve the false contour, 2 line memory is needed. Therefore, the proposed method can maximize the contrast image quality by having a minimal computational complexity. 6 Conclusion We have developed the image contrast enhancement algorithm that combines the optimized histogram equalization and the modified ED methods in order to improve the local brightness and the contouring artifacts together. We derived the sub-bloc size and step size of local histogram equalization that can be applied without causing the blocing artifacts for large-size LCD TV. This experimental results show that the proposed method preserves the local image brightness and suppresses the contouring artifacts, compared to the method using only local histogram equalization. The proposed module can be incorporated into the image processor that is used to control the images of LCD TV or the timing controller that provides the control signals for flat panel displays. Acnowledgment This wor has been supported by the MKE (The Ministry of Knowledge Economy), Korea, under the ITRC (Information Technology Research Center) support program supervised by the NIPA (National IT Industry Promotion Agency) (NIPA-2010-C1090-1001-0006). [4] S. M. Pizer et al, Adaptive histogram equalization and its variations, Computer Vision Graphics and Image Processing, Vol.39, pp.355 368, 1987. [5] J. Y. Ki L. S. Ki and S. H. Hwang, An advanced contrast enhancement using partially overlapped sub-bloc histogram equalization, IEEE Trans. on Circuits and Systems for Video Technology, Vol.11, No.4, pp.475-484, 2001. [6] Yeong-Taeg Ki Contrast enhancement using brightness preserving bi-histogram equalization, IEEE Trans. on Consumer Electronics, Vol.43, No.1, pp.1-8, 1997. [7] K. S. Si C. P. Tso, and Y. Y. Tan, Recursive sub-image histogram equalization applied to gray scale images, Pattern Recognition Letters, Vol.28, No.10, pp.1209-1221, 2007. [8] T. Liu, Probabilistic error diffusion for image enhancement, IEEE Trans. on Consumer Electronics, Vol. 53, No. 2, pp. 528-534, May 2007. [9] S.-J. Kang, H.-C. Do, and J.-H. Shin, Reduction of low gray-level contours using error diffusion based on emission characteristics of PDP, IEEE Trans. on Consumer Electronics, vol.50, No.2, May. 2004. [10] H. Chen, J. Sung, T. Ha, and Y. Par, Locally pixel-compensated baclight dimming on LED-baclit LCD TV Journal of the SID, 15/12, pp. 981-988, 2007. [11] T. K. Ki J. K. Pai, and B. S. Kang, Contrast enhancement system using spatially adaptive histogram equalization with temporal filtering, IEEE Trans. on Consumer Electronics, Vol. 44, No. 1, pp. 82-86, Feb. 1998. References: [1] R. C. Conzalez and R. E Woods, Digital Image Processing, New Jersey: Prentice-Hall, Inc., 2001. [2] S. D. Chen and A. Rahman Ramli, Contrast enhancement using recursive mean-separate histogram equalization for scalable brightness preservation, IEEE Trans. on Consumer Electronics, Vol.49, No.4, pp.1301-1309, 2003. [3] Z. Chen and B. R. Abidi, Gray-level grouping (GLG) : An automatic method for optimized image contrast enhancement-part I : The basic method, IEEE Trans. on Image Processing, Vol.15, No.8, pp.2290-2302, 2006. ISBN: 978-960-474-262-2 159