Multiclass Road Sign Detection using Multiplicative Kernel

Size: px
Start display at page:

Download "Multiclass Road Sign Detection using Multiplicative Kernel"

Transcription

1 Proceeding of the Croatian Computer Viion Workhop, Year 1 Multicla Road Sign Detection uing Multiplicative Kernel Valentina Zadrija Mireo d. d. Zagreb, Croatia valentina.zadrija@mireo.hr Siniša Šegvić Faculty of Electrical Engineering and Computing Univerity of Zagreb Zagreb, Croatia inia.egvic@fer.hr Abtract We conider the problem of multicla road ign detection uing a claification function with multiplicative kernel compried from two kernel. We how that problem of detection and within-foreground claification can be jointly olved by uing one kernel to meaure object-background difference and another one to account for within-cla variation. The main idea behind thi approach i that road ign from different foreground variation can hare feature that dicriminate them from background. The claification function training i accomplihed uing SVM, thu feature haring i obtained through upport vector haring. Training yield a family of linear detector, where each detector correpond to a pecific foreground training ample. The redundancy among detector i alleviated uing k- medoid clutering. Finally, we report detection and claification reult on a et of road ign image obtained from a camera on a moving vehicle. Keyword multicla object detection; object claification; road ign; SVM; multiplicative kernel; feature haring; clutering I. INTRODUCTION Road ign detection and claification i an exciting field of computer viion. There are variou application of the road ign detection and claification in driving aitance ytem, autonomou intelligent vehicle and automated traffic inventorie. The latter one i of particular interet to u ince traffic inventorie include periodical on-ite aement carried out by trained afety expert team. Current road condition i then manually compared againt the reference tate in the inventory. In current practice, the proce i tediou, error prone and cotly in term of expert time. Recently, there have been everal attempt to at leat partially automate the proce [1], [2]. Thi paper preent an attempt to partially automate thi proce in term of road ign detection and claification uing the multiplicative kernel. In general, detection and claification are typically two eparate procee. The mot common detection method i liding window approach, where each window in the original image i evaluated by a binary claifier in order to determine whether the window contain an object. When the propective object location are known, claification tage i applied only on the reulting window. One common claification approach Thi work ha been upported by reearch project Vita (EuropeAid/131920/M/ACT/HR) and Reearch Centre for Advanced Cooperative Sytem (EU FP7 #285939). include partitioning the object pace into ubclae and then training a dedicated claifier for each ubcla. Thi approach i known a one-veru-all claification. In thi paper, we focu on detection and claification of ideogram-baed road ign a a one-tage proce. We employ a jointly learned family of linear detector obtained through Support Vector Machine (SVM) learning with multiplicative kernel a preented in [3]. In it original form, SVM i a binary claifier, but multiplicative kernel formulation enable the multicla detection. Multiplicative kernel i defined a a product of two kernel, namely between-cla kernel k x and within-cla kernel k a decribed in Section IV. Betweencla kernel k x i dedicated for detection, i.e. the foregroundbackground claification. Within-cla kernel k i ued for within-foreground claification, i.e. to dicriminate between variou object ubclae. The reult of SVM training i a et of upport vector and correponding weight which are then ued to generate a family of detector. The detector are obtained by tuning within-cla tate into the multiplicative kernel a decribed in Section IV-B. The key point here i that all detector hare the ame upport vector, but weight of pecific upport vector vary depending on the within-cla tate value. We evaluate the decribed approach on a et of predefined road ign ubclae defined in Section III and preent experimental reult in Section V. II. RELATED WORK In recent year, a lot of work ha been propoed in the area of multicla object detection. We will addre the iue from both road ign detection apect a well a general multicla object detection and feature haring. The approach preented in [4] trie to olve multi-view face detection problem. Foreground cla i partitioned into ubclae according to variation in face orientation. For each ubcla, a correponding detector i learned. Thi approach exhibit everal problem when ued with a larger number of ubclae. More pecifically, number of feature, number of fale poitive and the total training time grow linearly with the number of ubclae. Poter Seion 37

2 Proceeding of the Croatian Computer Viion Workhop, Year 1 Fig. 1. Ditribution of ample with repect to road ign ubclae S = {1, 2, 3, 4, 5} for training dataet. Below each ubcla label v, repreentative ubcla member are hown. Road ign hown for ubcla v=1 are informal, i.e. ubcla contain 9 different road ign. Further, author in [5] focu on feature haring uing JointBoot procedure. In contrat to [4], where number of feature grow linearly with the number of ubclae, the author have experimentally hown that the number of feature grow logarithmically with repect to the number of ubclae. Additionally, the author howed that a jointly trained detector require a ignificantly maller number of feature in order to achieve the ame performance a a independent detector. The approach preented in [6] deal with problem of multiview detection uing the o called Vector Booted Tree procedure. At the detection time, the input i examined by a equence of node tarting from the root node. If the input i claified by the detector of current node a a member of the object cla, it i paed to the node children. Otherwie, it i rejected a a background. The drawback of thi approach i that it require the uer to predefine the tree tructure and chooe the poition to plit. Similar to [6], the approach preented in [7] alo employ a claifier tructured in a form of a tree. However, in contrat to [6], the tree claifier i contructed automatically - the node plit are achieved through unupervied clutering. The algorithm i iterative, i.e. at the beginning it tart with an empty tree and training ample which are aigned weight. By adding a node to the tree, the ample weight are modified accordingly. Additionally, if a plit i achieved, parent claifier of all node along the way to the root node are modified. The concept of feature haring i alo explored through hape-baed hierarchical compoitional model [8], [9], [10]. Thee model are ued for object categorization [8], [10], but alo for multi-view detection, [9]. Different object categorie can hare part or appearance. Part on lower hierarchy level are combined into larger part on higher level. In general, part on lower level are hared amongt variou object categorie, while thoe in higher level are more category pecific. Thi i imilar to approach employed in thi paper, however, in thi paper, the feature haring i obtained through a ingle-level upport vector haring. The approach preented in [1] decribe a road ign detection and recognition ytem baed on haring feature. The detection ubytem i a two tage proce compried of color-baed egmentation and SVM hape detection. The recognition ubytem comprie GentleBoot algorithm and Rotation-Scale-Tranlation Invariant template matching. III. PROBLEM DEFINITION In thi paper, we focu on detection and claification of ideogram-baed road ign. The cla of all road ign exhibit variou foreground variation with repect to the ign hape but alo on preence or abence of thick red rim and ideogram type. We aim to jointly train detector to dicriminate road ign from background a well a to produce foreground variation etimate, i.e. ubcla label. We will ue term "foreground variation" and "ubcla" interchangeably in the ret of the paper denoting the ame concept. Fig. 1 depict within-cla road ign ubclae which we aim to etimate. The cla of all road ign i compried out of five variation denoted with label v from S = {1, 2, 3, 4, 5}. Subcla v=1 include triangular warning ign with thick red. A mall ubet of the ubcla member i hown in Fig. 1, i.e. the ubcla contain nine different road ign. Thee ign belong to the category A according to the Vienna Convention [11]. Subcla v=2 contain circular "End of no overtaking zone" ign which belong to the category C of informative ign. On the other hand, member of ubcla v=3 "Priority road" and "End of Priority Road" are rhomb-haped. Subcla v=4 include quare-haped ign which are a ubet of informative category C road ign. Finally, the ubcla v=5 conatin circular "Speed Limit" ign characterized with the thick red rim which belong to the category B of prohibitory ign. We dicu the decribed road-ign variation in our dataet and the motivation for our approach a follow. Firt, we dicu motivation for partitioning road ign into ubclae according to the ditribution hown in Fig. 1. The dataet i extracted from the video recorded with camera mounted on top of a moving vehicle. Video equence are recorded at daytime, at different weather condition [2]. Further, a Fig. 1 how, the ditribution of ign in the dataet i unbalanced, i.e. certain variation like triangular ign are characterized with large number of intance, while ome other have a mall number of occurrence. In particular, the "End of Priority Road" ign hown a a part of the ubcla v=3 in Fig. 1 ha only nine intance in training dataet. In the approach where we build a ingle detector for a particular ubcla, it i clear that detector trained with only nine ample would have very poor detection rate. However, the "End of Priority Road" ign hare the ame hape a the "Priority Road" ign. If we were to group them into a ingle ubcla, we could exploit foreground-variation feature haring. Other heterogeneou ubclae are deigned with the ame motivation. Note that the ubcla v=5 i alo a heterogeneou ubcla, i.e. it contain variou peed limit ign which hare the thick red rim and the zero digit, ince the peed limit are uually multiple of ten. Second, according to the ubclae defined in Fig. 1, oberve that ign belonging to different ubclae alo hare imilaritie. For example, the "End of no overtaking zone" (ubcla v=2) ign and "End of Priority Road" (ubcla v=3) both hare the ame ditinctive croover mark. Thi imilarity could improve dicrimination capability of both ign with repect to the background cla. Thi ugget that if we olve detection and claification problem for all ubclae together, we could benefit from within-cla feature haring. Poter Seion 38

3 Proceeding of the Croatian Computer Viion Workhop, Year 1 Therefore, due to the decribed characteritic of the dataet ditribution, a well a the nature of ign imilaritie, we decided to employ a method preented in [3] where a claification function i learned jointly for all within-cla variation. The aim of thi approach i to form the claification function which could exploit the fact that different variation hare feature againt background, but at the ame time provide within-cla dicrimination. IV. DETECTION AND CLASSIFICATION APPROACH The overall road ign detection and claification proce i hown in Fig. 2. The detailed decription i a follow. For a given feature vector x n computed for an image patch, the goal i to decide whether it repreent an intance of a road ign cla and, if o, to produce the correponding ubcla etimate v from S = {1, 2, 3, 4, 5}. Let x i denote feature vector of the i-th road ign training ample belonging to the ubcla with label v i. The feature vector are given a HOG feature [15]. According to the above defined parameter, the claification function C(x, i defined a follow: Fig. 2. Training and detector contruction outline the claification function. In thi way, we achieve the withincla feature haring a well a the within-cla dicrimination. 0, x i foreground from theame ubcla a x (, ) i A. Claification Function Training C x i 0, otherwie Thi correpond to the non-parametric approach preented in [3]. The parametric approach [3] i impler, however, it require each ubcla v to be decribed with a pecific parameter. The role of parameter i to decribe member of the ubcla in a unique way. In multiview domain, the parameter typically correpond to the view angle or object poe. However, in the road ign domain the ubclae are heterogeneou and cannot be decribed with a ingle parameter. For thi reaon, the claification function employ the foreground ample feature vector x i in order to decribe a pecific ubcla. Since the road ign ubclae are deigned with the goal that ign within the ubcla are imilar, it i to be expected that they will alo exhibit imilar feature vector value x i. SV Parameter denote the Lagrange multiplier of the -th upport vector [12], x the particular upport vector, x i the i-th foreground training ample, while k (x, x i ) denote the withincla kernel and k x (x, x i ) the between-cla kernel. In addition, the product of firt two term in equation (2) can be ummarized into a ingle term '(, which denote the weight of the -th upport vector x for foreground ample x i : i '( k ( x, x ) into (2) and (3) A a reult, the upport vector for which the within-cla kernel k yield higher value will have a larger influence on i x The claification function (2) training i achieved uing SVM. The training ample take form of tuple (x,. Each foreground training ample x i aigned it correponding ample index i. Background training ample x are obtained from image patche without road ign. Each background training ample x can be aociated with any index of a foreground training ample in order to form a valid tuple. More pecifically, background ample x i a negative with repect to all foreground ample. The number of uch combination i huge and correpond to #( NB) #( NF) The parameter #(NB) correpond to the total number of background and #(NF) to the total number of foreground training ample. Due to combinatorial complexity, including all negative ample in SVM training would not be practical. In order to atify requirement of within-cla feature Therefore, the boottrap training i employed a a hard haring againt the background cla, a well a within-cla negative mining technique. In boottrap training, only #(NB) dicrimination, the claification function C(x, i repreented negative are initially included in training. Thee ample are a a product of two kernel: aigned foreground ample indice in a random fahion. After each training round, all negative ample are evaluated by the C( x, k ( x, x ) k ( x, x) claification function (2). Fale poitive are added to the negative et and the SVM training i repeated. Thi i an iterative proce which converge when there are no more fale poitive to add. B. Individual Detector Contruction C(x, i learned a a function of a foreground variation parameter i, rather than learning eparate detector for each i. Individual detector w(x, are obtained from the claification function by plugging in pecific foreground ample value x i w(x, '( kx( x, x) SV Poter Seion 39

4 Proceeding of the Croatian Computer Viion Workhop, Year 1 Note that with fixed foreground variation i and known et of upport vector x, we can precompute the within-cla kernel value k (x, x i ) and conequently upport vector weight '(. Therefore, the within-cla kernel k i not evaluated at detection time. Rather, at detection time, we only evaluate the between-cla kernel k x. Thi fact affect our choice for withincla and between-cla kernel. Firt, we dicu within-cla kernel k. Road ign ubclae are difficult to eparate and i therefore important for k to be able to eparate nonlinear problem. Therefore, Gauian RBF kernel wa choen for that purpoe w(x, SV yield the bet ilhouette value. ' T ( ) ( x x) w( x V. EXPERIMENTAL RESULTS In thi way, the detection i achieved by applying a imple dot product between the image patch and the detector weight denoted a w(. Note that all detector hare the ame et of upport vector. In thi way, feature haring among variou detector i achieved. C. Detection Approach In the detection proce, we employ the well known liding window technique. Each window i evaluated by a family of linear detector w(x, contructed according to (7). From all detector repone, the one with maximum value i choen a a reult. If thi value i poitive, the window i claified a a road ign belonging to the ubcla of x i. Otherwie, the window i dicarded a a background. Note that thi i imilar to the k-nearet neighbor (k-nn) method, with parameter k=1, i.e. the object i imply aigned to the cla of the nearet neighbor elected among all detector [13]. However, in the 1-NN approach, the number of evaluated detector i ignificant, i.e. correpond to the number of foreground ample #(NF). Evaluating all #(NF) detector at the detection tage would make the detection extremely low. In addition, ince foreground ample belonging to the ame ubcla are imilar, there may be redundancy among detector. In order to identify a repreentative et from a family of total #(NF) detector, we ue the k-medoid clutering technique. The k-medoid technique i choen due to it implicity and alo becaue it i le prone to outlier influence than, for intance, k-mean method. Clutering yield a et of k < #(NF) medoid which are then ued in the detection phae. In clutering, each detector w(x, i repreented with a vector of it upport vector weight: ' ' ' 1 2 SV i '( (, (,, ( ) where '(, {1,..., SV} denote the particular upport weight defined with (3), while SV denote total number of upport vector. A a ditance meaure, we ued Dit ( i, j) defined a follow: T ' ( ' ( j) Dit ( i, j) 1 ' ( ' ( j) The appropriate number of medoid k i choen in an k ( x, xi ) exp( D( x, xi )) iterative proce, where we decreae number of medoid gradually and meaure the clutering quality. Initially, the where D(x,x i ) denote Euclidian ditance and target number of medoid, i.e. cluter center k i et to 50% of correponding parameter. Due to the fact that RBF i evaluated only during training and detector contruction, thi choice doen't impoe performance penalty during detection. the initial number of detector, i.e. foreground ample. Thi number wa choen by a rule of thumb. Then, we apply clutering according to the choen number of medoid k. In order to meaure clutering quality, we compute correponding Secondly, ince the between-cla kernel k x i evaluated ilhouette value [14]. The ilhouette value provide an etimate during detection, it i important for k x to be fat. Therefore, we of how the well obtained medoid repreent the data in their choe linear kernel for that purpoe. By ubtituting the linear kernel formulation k x (x,x i ) = x T correponding cluter. In each iteration, the target number of i x into (5) we obtain the final cluter i decreaed by a certain factor and the above proce form of our detector: i repeated. Final clutering outline i choen a the one which In thi ection, we decribe the evaluation of the above decribed method according to the foreground variation ditribution preented in Section III. In our experiment, we ued three dijoint dataet, decribed below [2]. A we already treed, the ditribution hown in Fig. 1 i unbalanced, i.e. ubcla v=1 contain at leat three time more ample that other clae. In uch unbalanced dataet, there i a poibility for pecific road ign variation with maller number of ample to be treated a a noie, e.g. ubcla v=3. In order to tet thi hypothei, we experimented with the number of ample per cla. Namely, we oberved detection performance for N POS =200, 300 and 400 ample per ubcla. For the ubclae with the number of ample lower than N POS, we imply ue all the ample for the ubcla, e.g. ubclae v=2, v=3 and N POS =400. The training ample are extracted randomly from training dataet pool compriing 2153 road ign. Negative dataet contain 4000 image patche extracted randomly from image of background outdoor cene. In order to monitor clutering performance, we ue the tet dataet compried out of 3000 cropped image. Detail of the tet dataet are given in Section V-B. A. Training Approach The training of detector i achieved a decribed in Section IV. A feature, we ued HOG vector [15] computed from training image cropped and caled to 24 x 24 pixel. Cell ize i et to 4 pixel, where each cell i normalized within a block of four cell. In order to increae performance, we ue block overlapping with block tride et to ize of ingle cell. The training i achieved uing SVMlight [16] with multiplicative kernel. In contrat to [3], where parameter of the RBF kernel Poter Seion 40

5 Proceeding of the Croatian Computer Viion Workhop, Year 1 (a) Fig. 3. Example of a detection and claification: (a) correct claification, (b) correct claification and fale poitive. (6) i et to a fixed value, we perform cro validation on training dataet in each boottrapping round in order to obtain the bet value. We compared both training approache and the one with optimized value yield a lower number of upport vector. Thi ugget a better mapping in the tranformed feature pace. More pecifically, with training et compried out of 1325 poitive ample and 4000 negative ample, the training with the fixed value yield a et of 857 upport vector. On the other hand, the training with the optimized decreae the number of upport vector for 10%. The training yield a family of #(NF) detector. The #(NF) correpond to the number of foreground training ample which depend on the number of training ample per ubcla N POS, Table I. Due to performance reaon, we ue k-medoid clutering in order to elect a repreentative et of detector from total #(NF) detector. The clutering i implemented in Matlab according to the Partitioning Around Medoid method [17]. In each clutering iteration, we monitor the ilhouette value and the validation reult on the tet dataet compried out of cropped image. Note that thi dataet i dijoint from tet dataet ued for detection. Interetingly, better ilhouette value correpond to a maller number fale negative obtained from validation on tet data. The reult of clutering are ummarized in Table I. depending on the N POS and #(NF). Reulting number of detector k correpond approximately to 30% of total training ample #(NF). Lower k value lead to poor validation reult and mall ilhouette value. B. Detection and Claification Reult The tet dataet for detection evaluation contain 1038 image in 720x576 reolution. From the total 1038 image, we elected 200 image and ued them for performance evaluation. In thee image, there were 214 phyical road ign. Table II. how the detection and foreground etimation reult for the three cae tudie. We report the detection rate D, the claification rate C, the fale poitive rate FP and the fale poitive rate per image FP/I. D i defined a a number of detected ign with repect to the total number of ign, while C and FP correpond to the number of correct claification and fale detection with repect to the total number of ign, repectively. FP/I correpond to the number of fale detection with repect to the number of image. Column denoted with ign how difference in above metric with repect to configuration denoted by N POS =200. We tarted the experiment with N POS =200 ample per ubcla. Thi configuration achieve overall detection rate of (b) 90% with fale poitive rate of 43%. Next, N POS =300 achieve 4% rie in detection rate giving total 94%, and 3% rie in claification rate, i.e. 93%. However, it i characterized with a large fale poitive rate of 55%. In liding window detection, ome computer viion librarie like OpenCV [18] employ fale poitive detection policy where an object mut exhibit at leat n detection in order to be accounted a a reult. Thi i undertandable, ince liding window technique exhibit multiple repone around ingle object. In thi work, we didn't experiment with thi property, however we believe that it would decreae fale poitive rate. Finally, the configuration N POS =400 exhibit wore reult with repect to N POS =300. Thi upport our hypothei that the unbalanced dataet N POS =400 treat ubclae with a lower number of ample a a noie giving the lower overall detection rate. Table III. depict D and C rate, a well a the FP rate per pecific ubcla v for configuration N POS =300 and N POS =200, repectively. The ubcla v=1 achieve better reult when a larger number of ample i ued. Thi i undertandable, ince thi ubcla comprie a large number of different road ign. Interetingly, the ubcla v=3 which ha only 98 ample (8% of the total ample for ubcla v=1) achieve detection and claification rate of 100% in all cae tudie. Subclae v=2 and v=4 achieve lower detection and claification rate with repect to other ubclae. Subcla v=2 i circle-haped but lack red rim in order to hare feature with ubcla v=5. On the other hand, ubcla v=4 i rectangle-haped and gain le benefit from feature haring with other ubclae. Subcla v=5 obtain imilar reult for N POS =300 and N POS =200. FP ditribution per ubcla for N POS =200 how that ubcla v=4 exhibit the larget number of fale poitive, i.e. 58%. Example of fale poitive claified a member of ubcla v=4 include building window. On the other hand, N POS =300 yield a rather balanced FP ditribution, where ubclae v=1, 4 and 5 obtain FP rate of approximately 30%. Example of detection and claification are given in Fig. 3a and Fig. 3b. Fig. 3a illutrate an example of a correct claification, where "Speed Limit" ign i claified a a member of ubcla v=5 (orange dotted line), while the "Children" ign i claified a a member of ubcla v=1 (green dotted line). Fig. 3.b illutrate correct claification, a well a two fale poitive. The "Priority Road" ign i correctly claified a a ubcla v=3 (cyan dotted line). The "Weight Limit" ign wa not preent in training data, however, due to imilarity with "Speed Limit" ign, it i claified a a member of ubcla v=5 (orange dotted line). The latter one indicate within-cla feature haring. The triangle-like object i incorrectly claified a a member of ubcla v=1 (green dotted line). VI. CONCLUSION In thi paper, we conidered a road-ign detection technique baed on a multiplicative kernel. One of the major challenge wa a poorly balanced dataet, where triangular warning ign have at leat three time more intance than other ubclae. Our approach i baed on a premie that different ign ubclae hare feature which dicriminate them from background. Therefore, intead of learning a dedicated detector for each ubcla, we trained ingle claification function for all ubclae uing SVM [3]. Individual detector Poter Seion 41

6 Proceeding of the Croatian Computer Viion Workhop, Year 1 TABLE I. CLUSTERING RESULTS N POS #(NF) k TABLE II. DETECTION AND CLASSIFICATION N POS D D C C FP FP FP/I FP/I % - 90% - 45% - 47% % +3% 93% +3% 55% +10% 58% +11% % -1% 90% 0% 43% -2% 45% -3% TABLE III. RESULTS PER SUBCLASS N POS = 300 N POS = 200 v D C FP D C FP 1 100% 100% 34% 91% 91% 15% 2 82% 77% 0% 82% 82% 0% 3 100% 100% 4% 100% 100% 5% 4 88% 82% 31% 76% 71% 58% 5 92% 92% 31% 93% 93% 22% are afterward contructed from a hared et of upport vector. Major benefit of uch approach with repect to eparately trained detector lie in feature haring which enhance the detection rate for ubclae with lower number of ample. In comparion to partition baed approache, thi approach doe not require the training ample to be labeled with ubcla parameter in order to learn claification function. Thi fact ha proven to be ueful for our domain, ince the road ign ubclae are heterogeneou and it i hard to decribe a ubcla with a ingle parameter. Intead, each training ample i labeled with it correponding HOG feature vector. In thi way we obtain #(NF) ubcla parameter and conequently #(NF) detector from the claification function, where #(NF) correpond to the number of foreground ample. However, due to performance iue, we perform clutering in order to obtain a repreentative et of detector from a et of total #(NF) detector. The reduced et of detector i ued in detection. Uing the decribed method, we achieved the bet detection rate of 94% at a relatively high fale poitive rate of 55%. We experimented with different number of ample per ubcla in order to oberve the effect on detection rate. The obtained reult howed that detector trained on a limited number of ample, i.e. 300 ample per cla obtain better detection reult with repect to the larger number of ample. Due to the fact that thi method ha hown promiing reult in road ign domain, in future work we plan to explore it applicability in the domain of multiview vehicle detection. ACKNOWLEDGMENT The author wih to thank Joip Krapac for ueful uggetion on early verion of thi paper. REFERENCES [1] J.-Y. Wu, C.-C. Teng, C.-H. Chang, J.-J. Lien, J.-C. Chen, and C. T. Tu, Road ign recognition ytem baed on GentleBoot with haring feature, in Sytem Science and Engineering (ICSSE), 2011 International Conference on, 2011, pp [2] S. Šegvić, K. Brkić, Z. Kalafatić, and A. Pinz, Exploiting temporal and patial contraint in traffic ign detection from a moving vehicle, Machine Viion and Application, pp. 1 17, [Online]. Available: [3] Q. Yuan, A. Thangali, V. Ablavky, and S. Sclaroff, Learning a family of detector via multiplicative kernel, Pattern Analyi and Machine Intelligence, IEEE Tranaction on, vol. 33, no. 3, pp , [4] P. Viola and M. Jone, Fat Multi-View Face Detection, in Proc. of IEEE Conf. Computer Viion and Pattern Recognition, [5] A. Torralba, K. Murphy, and W. Freeman, Sharing viual feature for multicla and multiview object detection, Pattern Analyi and Machine Intelligence, IEEE Tranaction on, vol. 29, no. 5, pp , [6] C. Huang, H. Ai, Y. Li, and S. Lao, High-performance rotation invariant multiview face detection, Pattern Analyi and Machine Intelligence, IEEE Tranaction on, vol. 29, no. 4, pp , [7] B. Wu and R. Nevatia, Cluter booted tree claifier for multi-view, multi-poe object detection, in Computer Viion, ICCV IEEE 11th International Conference on, 2007, pp [8] S. Fidler and A. Leonardi, Toward calable repreentation of object categorie: Learning a hierarchy of part, in Computer Viion and Pattern Recognition, CVPR 07. IEEE Conference on, 2007, pp [9] L. Zhu, Y. Chen, A. Torralba, W. Freeman, and A. Yuille, Part and appearance haring: Recurive compoitional model for multiview, in Computer Viion and Pattern Recognition (CVPR), 2010 IEEE Conference on, 2010, pp [10] Z. Si and S.-C. Zhu, Unupervied learning of tochatic and-or template for object modeling, in ICCV Workhop, 2011, pp [11] Inland tranport comitee, Convention on road ign and ignal, Economic comiion for Europe, [12] C. M. Bihop, Pattern Recognition and Machine Learning (Information Science and Statitic). Secaucu, NJ, USA: Springer-Verlag New York, Inc., [13] D. Bremner, E. Demaine, J. Erickon, J. Iacono, S. Langerman, P. Morin, and G. Touaint, Output-enitive algorithm for computing nearet-neighbour deciion boundarie, in Algorithm and Data Structure, er. Lecture Note in Computer Science, F. Dehne, J.-R. Sack, and M. Smid, Ed. Springer Berlin Heidelberg, 2003, vol. 2748, pp [Online]. Available: 39 [14] P. Roueeuw, Silhouette: a graphical aid to the interpretation and validation of cluter analyi, J. Comput. Appl. Math., vol. 20, no. 1, pp , Nov [Online]. Available: (87) [15] N. Dalal and B. Trigg, Hitogram of oriented gradient for human detection, in Computer Viion and Pattern Recognition, CVPR IEEE Computer Society Conference on, vol. 1, 2005, pp vol. 1. [16] T. Joachim, Advance in kernel method, B. Scholkopf, C. J. C. Burge, and A. J. Smola, Ed. Cambridge, MA, USA: MIT Pre, 1999, ch. Making large-cale upport vector machine learning practical, pp [Online]. Available: [17] S. Theodoridi and K. Koutroumba, Pattern Recognition, Fourth Edition, 4th ed. Academic Pre, [18] G. Bradki, The OpenCV Library, Dr. Dobb Journal of Software Tool, Poter Seion 42

Performance of a Robust Filter-based Approach for Contour Detection in Wireless Sensor Networks

Performance of a Robust Filter-based Approach for Contour Detection in Wireless Sensor Networks Performance of a Robut Filter-baed Approach for Contour Detection in Wirele Senor Network Hadi Alati, William A. Armtrong, Jr., and Ai Naipuri Department of Electrical and Computer Engineering The Univerity

More information

Universität Augsburg. Institut für Informatik. Approximating Optimal Visual Sensor Placement. E. Hörster, R. Lienhart.

Universität Augsburg. Institut für Informatik. Approximating Optimal Visual Sensor Placement. E. Hörster, R. Lienhart. Univerität Augburg à ÊÇÅÍÆ ËÀǼ Approximating Optimal Viual Senor Placement E. Hörter, R. Lienhart Report 2006-01 Januar 2006 Intitut für Informatik D-86135 Augburg Copyright c E. Hörter, R. Lienhart Intitut

More information

A TOPSIS based Method for Gene Selection for Cancer Classification

A TOPSIS based Method for Gene Selection for Cancer Classification Volume 67 No17, April 2013 A TOPSIS baed Method for Gene Selection for Cancer Claification IMAbd-El Fattah,WIKhedr, KMSallam, 1 Department of Statitic, 3 Department of Deciion upport, 2 Department of information

More information

MAT 155: Describing, Exploring, and Comparing Data Page 1 of NotesCh2-3.doc

MAT 155: Describing, Exploring, and Comparing Data Page 1 of NotesCh2-3.doc MAT 155: Decribing, Exploring, and Comparing Data Page 1 of 8 001-oteCh-3.doc ote for Chapter Summarizing and Graphing Data Chapter 3 Decribing, Exploring, and Comparing Data Frequency Ditribution, Graphic

More information

Trainable Context Model for Multiscale Segmentation

Trainable Context Model for Multiscale Segmentation Trainable Context Model for Multicale Segmentation Hui Cheng and Charle A. Bouman School of Electrical and Computer Engineering Purdue Univerity Wet Lafayette, IN 47907-1285 {hui, bouman}@ ecn.purdue.edu

More information

3D SMAP Algorithm. April 11, 2012

3D SMAP Algorithm. April 11, 2012 3D SMAP Algorithm April 11, 2012 Baed on the original SMAP paper [1]. Thi report extend the tructure of MSRF into 3D. The prior ditribution i modified to atify the MRF property. In addition, an iterative

More information

A Novel Feature Line Segment Approach for Pattern Classification

A Novel Feature Line Segment Approach for Pattern Classification 12th International Conference on Information Fuion Seattle, WA, USA, July 6-9, 2009 A Novel Feature Line Segment Approach for Pattern Claification Yi Yang Intitute of Integrated Automation Xi an Jiaotong

More information

A User-Attention Based Focus Detection Framework and Its Applications

A User-Attention Based Focus Detection Framework and Its Applications A Uer-Attention Baed Focu Detection Framework and It Application Chia-Chiang Ho, Wen-Huang Cheng, Ting-Jian Pan, Ja-Ling Wu Communication and Multimedia Laboratory, Department of Computer Science and Information

More information

A New Approach for In-Vehicle Camera Traffic Sign Detection and Recognition.

A New Approach for In-Vehicle Camera Traffic Sign Detection and Recognition. A New Approach for In-Vehicle Camera Traffic Sign Detection and Recognition. Andrzej Ruta, Yongmin Li School of Information Sytem, Computing & Mathematic Brunel Univerity Uxbridge, Middleex UB8 3PH, United

More information

IMPROVED JPEG DECOMPRESSION OF DOCUMENT IMAGES BASED ON IMAGE SEGMENTATION. Tak-Shing Wong, Charles A. Bouman, and Ilya Pollak

IMPROVED JPEG DECOMPRESSION OF DOCUMENT IMAGES BASED ON IMAGE SEGMENTATION. Tak-Shing Wong, Charles A. Bouman, and Ilya Pollak IMPROVED DECOMPRESSION OF DOCUMENT IMAGES BASED ON IMAGE SEGMENTATION Tak-Shing Wong, Charle A. Bouman, and Ilya Pollak School of Electrical and Computer Engineering Purdue Univerity ABSTRACT We propoe

More information

Analyzing Hydra Historical Statistics Part 2

Analyzing Hydra Historical Statistics Part 2 Analyzing Hydra Hitorical Statitic Part Fabio Maimo Ottaviani EPV Technologie White paper 5 hnode HSM Hitorical Record The hnode i the hierarchical data torage management node and ha to perform all the

More information

Laboratory Exercise 6

Laboratory Exercise 6 Laboratory Exercie 6 Adder, Subtractor, and Multiplier The purpoe of thi exercie i to examine arithmetic circuit that add, ubtract, and multiply number. Each type of circuit will be implemented in two

More information

On successive packing approach to multidimensional (M-D) interleaving

On successive packing approach to multidimensional (M-D) interleaving On ucceive packing approach to multidimenional (M-D) interleaving Xi Min Zhang Yun Q. hi ankar Bau Abtract We propoe an interleaving cheme for multidimenional (M-D) interleaving. To achieved by uing a

More information

Comparison of Methods for Horizon Line Detection in Sea Images

Comparison of Methods for Horizon Line Detection in Sea Images Comparion of Method for Horizon Line Detection in Sea Image Tzvika Libe Evgeny Gerhikov and Samuel Koolapov Department of Electrical Engineering Braude Academic College of Engineering Karmiel 2982 Irael

More information

CENTER-POINT MODEL OF DEFORMABLE SURFACE

CENTER-POINT MODEL OF DEFORMABLE SURFACE CENTER-POINT MODEL OF DEFORMABLE SURFACE Piotr M. Szczypinki Iintitute of Electronic, Technical Univerity of Lodz, Poland Abtract: Key word: Center-point model of deformable urface for egmentation of 3D

More information

On combining Learning Vector Quantization and the Bayesian classifiers for natural textured images

On combining Learning Vector Quantization and the Bayesian classifiers for natural textured images On combining Learning Vector Quantization and the Bayeian claifier for natural textured image María Guiarro Dept. Ingeniería del Software e Inteligencia Artificial Facultad Informática Univeridad Complutene

More information

Image authentication and tamper detection using fragile watermarking in spatial domain

Image authentication and tamper detection using fragile watermarking in spatial domain International Journal of Advanced Reearch in Computer Engineering & Technology (IJARCET) Volume 6, Iue 7, July 2017, ISSN: 2278 1323 Image authentication and tamper detection uing fragile watermarking

More information

A New Approach for In-Vehicle Camera Traffic Sign Detection and Recognition

A New Approach for In-Vehicle Camera Traffic Sign Detection and Recognition MITSUBISHI ELECTRIC RESEARCH LABORATORIES http://www.merl.com A New Approach for In-Vehicle Camera Traffic Sign Detection and Recognition Andrzej Ruta, Fatih Porikli, Yongmin Li, Shintaro Watanabe, Hirohi

More information

Service and Network Management Interworking in Future Wireless Systems

Service and Network Management Interworking in Future Wireless Systems Service and Network Management Interworking in Future Wirele Sytem V. Tountopoulo V. Stavroulaki P. Demeticha N. Mitrou and M. Theologou National Technical Univerity of Athen Department of Electrical Engineering

More information

UC Berkeley International Conference on GIScience Short Paper Proceedings

UC Berkeley International Conference on GIScience Short Paper Proceedings UC Berkeley International Conference on GIScience Short Paper Proceeding Title A novel method for probabilitic coverage etimation of enor network baed on 3D vector repreentation in complex urban environment

More information

Gray-level histogram. Intensity (grey-level) transformation, or mapping. Use of intensity transformations:

Gray-level histogram. Intensity (grey-level) transformation, or mapping. Use of intensity transformations: Faculty of Informatic Eötvö Loránd Univerity Budapet, Hungary Lecture : Intenity Tranformation Image enhancement by point proceing Spatial domain and frequency domain method Baic Algorithm for Digital

More information

DAROS: Distributed User-Server Assignment And Replication For Online Social Networking Applications

DAROS: Distributed User-Server Assignment And Replication For Online Social Networking Applications DAROS: Ditributed Uer-Server Aignment And Replication For Online Social Networking Application Thuan Duong-Ba School of EECS Oregon State Univerity Corvalli, OR 97330, USA Email: duongba@eec.oregontate.edu

More information

3D MODELLING WITH LINEAR APPROACHES USING GEOMETRIC PRIMITIVES

3D MODELLING WITH LINEAR APPROACHES USING GEOMETRIC PRIMITIVES MAKARA, TEKNOLOGI, VOL. 9, NO., APRIL 5: 3-35 3D MODELLING WITH LINEAR APPROACHES USING GEOMETRIC PRIMITIVES Mochammad Zulianyah Informatic Engineering, Faculty of Engineering, ARS International Univerity,

More information

Building a Compact On-line MRF Recognizer for Large Character Set using Structured Dictionary Representation and Vector Quantization Technique

Building a Compact On-line MRF Recognizer for Large Character Set using Structured Dictionary Representation and Vector Quantization Technique 202 International Conference on Frontier in Handwriting Recognition Building a Compact On-line MRF Recognizer for Large Character Set uing Structured Dictionary Repreentation and Vector Quantization Technique

More information

Hassan Ghaziri AUB, OSB Beirut, Lebanon Key words Competitive self-organizing maps, Meta-heuristics, Vehicle routing problem,

Hassan Ghaziri AUB, OSB Beirut, Lebanon Key words Competitive self-organizing maps, Meta-heuristics, Vehicle routing problem, COMPETITIVE PROBABIISTIC SEF-ORGANIZING MAPS FOR ROUTING PROBEMS Haan Ghaziri AUB, OSB Beirut, ebanon ghaziri@aub.edu.lb Abtract In thi paper, we have applied the concept of the elf-organizing map (SOM)

More information

A DIVISIVE HIERARCHICAL CLUSTERING- BASED METHOD FOR INDEXING IMAGE INFORMATION

A DIVISIVE HIERARCHICAL CLUSTERING- BASED METHOD FOR INDEXING IMAGE INFORMATION A DIVISIVE HIERARCHICAL CLUSTERING- BASED METHOD FOR INDEXING IMAGE INFORMATION ABSTRACT Najva Izadpanah Department of Computer Engineering, Ilamic Azad Univerity, Qazvin Branch, Qazvin, Iran In mot practical

More information

CSE 250B Assignment 4 Report

CSE 250B Assignment 4 Report CSE 250B Aignment 4 Report March 24, 2012 Yuncong Chen yuncong@c.ucd.edu Pengfei Chen pec008@ucd.edu Yang Liu yal060@c.ucd.edu Abtract In thi project, we implemented the recurive autoencoder (RAE) a decribed

More information

SLA Adaptation for Service Overlay Networks

SLA Adaptation for Service Overlay Networks SLA Adaptation for Service Overlay Network Con Tran 1, Zbigniew Dziong 1, and Michal Pióro 2 1 Department of Electrical Engineering, École de Technologie Supérieure, Univerity of Quebec, Montréal, Canada

More information

Computer Aided Drafting, Design and Manufacturing Volume 25, Number 3, September 2015, Page 10

Computer Aided Drafting, Design and Manufacturing Volume 25, Number 3, September 2015, Page 10 Computer Aided Drafting, Deign and Manufacturing Volume 5, umber 3, September 015, Page 10 CADDM Reearch of atural Geture Recognition and Interactive Technology Compatible with YCbCr and SV Color Space

More information

else end while End References

else end while End References 621-630. [RM89] [SK76] Roenfeld, A. and Melter, R. A., Digital geometry, The Mathematical Intelligencer, vol. 11, No. 3, 1989, pp. 69-72. Sklanky, J. and Kibler, D. F., A theory of nonuniformly digitized

More information

AN ALGORITHM FOR RESTRICTED NORMAL FORM TO SOLVE DUAL TYPE NON-CANONICAL LINEAR FRACTIONAL PROGRAMMING PROBLEM

AN ALGORITHM FOR RESTRICTED NORMAL FORM TO SOLVE DUAL TYPE NON-CANONICAL LINEAR FRACTIONAL PROGRAMMING PROBLEM RAC Univerity Journal, Vol IV, No, 7, pp 87-9 AN ALGORITHM FOR RESTRICTED NORMAL FORM TO SOLVE DUAL TYPE NON-CANONICAL LINEAR FRACTIONAL PROGRAMMING PROLEM Mozzem Hoain Department of Mathematic Ghior Govt

More information

Edits in Xylia Validity Preserving Editing of XML Documents

Edits in Xylia Validity Preserving Editing of XML Documents dit in Xylia Validity Preerving diting of XML Document Pouria Shaker, Theodore S. Norvell, and Denni K. Peter Faculty of ngineering and Applied Science, Memorial Univerity of Newfoundland, St. John, NFLD,

More information

Implementation of a momentum-based distance metric for motion graphs. Student: Alessandro Di Domenico (st.no ), Supervisor: Nicolas Pronost

Implementation of a momentum-based distance metric for motion graphs. Student: Alessandro Di Domenico (st.no ), Supervisor: Nicolas Pronost Implementation of a momentum-baed ditance metric for motion graph Student: Aleandro Di Domenico (t.no 3775682), Supervior: Nicola Pronot April 3, 2014 Abtract Thi report preent the procedure and reult

More information

A Multi-objective Genetic Algorithm for Reliability Optimization Problem

A Multi-objective Genetic Algorithm for Reliability Optimization Problem International Journal of Performability Engineering, Vol. 5, No. 3, April 2009, pp. 227-234. RAMS Conultant Printed in India A Multi-objective Genetic Algorithm for Reliability Optimization Problem AMAR

More information

Markov Random Fields in Image Segmentation

Markov Random Fields in Image Segmentation Preented at SSIP 2011, Szeged, Hungary Markov Random Field in Image Segmentation Zoltan Kato Image Proceing & Computer Graphic Dept. Univerity of Szeged Hungary Zoltan Kato: Markov Random Field in Image

More information

Operational Semantics Class notes for a lecture given by Mooly Sagiv Tel Aviv University 24/5/2007 By Roy Ganor and Uri Juhasz

Operational Semantics Class notes for a lecture given by Mooly Sagiv Tel Aviv University 24/5/2007 By Roy Ganor and Uri Juhasz Operational emantic Page Operational emantic Cla note for a lecture given by Mooly agiv Tel Aviv Univerity 4/5/7 By Roy Ganor and Uri Juhaz Reference emantic with Application, H. Nielon and F. Nielon,

More information

A METHOD OF REAL-TIME NURBS INTERPOLATION WITH CONFINED CHORD ERROR FOR CNC SYSTEMS

A METHOD OF REAL-TIME NURBS INTERPOLATION WITH CONFINED CHORD ERROR FOR CNC SYSTEMS Vietnam Journal of Science and Technology 55 (5) (017) 650-657 DOI: 10.1565/55-518/55/5/906 A METHOD OF REAL-TIME NURBS INTERPOLATION WITH CONFINED CHORD ERROR FOR CNC SYSTEMS Nguyen Huu Quang *, Banh

More information

Planning of scooping position and approach path for loading operation by wheel loader

Planning of scooping position and approach path for loading operation by wheel loader 22 nd International Sympoium on Automation and Robotic in Contruction ISARC 25 - September 11-14, 25, Ferrara (Italy) 1 Planning of cooping poition and approach path for loading operation by wheel loader

More information

Multi-Target Tracking In Clutter

Multi-Target Tracking In Clutter Multi-Target Tracking In Clutter John N. Sander-Reed, Mary Jo Duncan, W.B. Boucher, W. Michael Dimmler, Shawn O Keefe ABSTRACT A high frame rate (0 Hz), multi-target, video tracker ha been developed and

More information

Refining SIRAP with a Dedicated Resource Ceiling for Self-Blocking

Refining SIRAP with a Dedicated Resource Ceiling for Self-Blocking Refining SIRAP with a Dedicated Reource Ceiling for Self-Blocking Mori Behnam, Thoma Nolte Mälardalen Real-Time Reearch Centre P.O. Box 883, SE-721 23 Väterå, Sweden {mori.behnam,thoma.nolte}@mdh.e ABSTRACT

More information

Motion Control (wheeled robots)

Motion Control (wheeled robots) 3 Motion Control (wheeled robot) Requirement for Motion Control Kinematic / dynamic model of the robot Model of the interaction between the wheel and the ground Definition of required motion -> peed control,

More information

A PROBABILISTIC NOTION OF CAMERA GEOMETRY: CALIBRATED VS. UNCALIBRATED

A PROBABILISTIC NOTION OF CAMERA GEOMETRY: CALIBRATED VS. UNCALIBRATED A PROBABILISTIC NOTION OF CAMERA GEOMETRY: CALIBRATED VS. UNCALIBRATED Jutin Domke and Yianni Aloimono Computational Viion Laboratory, Center for Automation Reearch Univerity of Maryland College Park,

More information

[N309] Feedforward Active Noise Control Systems with Online Secondary Path Modeling. Muhammad Tahir Akhtar, Masahide Abe, and Masayuki Kawamata

[N309] Feedforward Active Noise Control Systems with Online Secondary Path Modeling. Muhammad Tahir Akhtar, Masahide Abe, and Masayuki Kawamata he 32nd International Congre and Expoition on Noie Control Engineering Jeju International Convention Center, Seogwipo, Korea, Augut 25-28, 2003 [N309] Feedforward Active Noie Control Sytem with Online

More information

How to Select Measurement Points in Access Point Localization

How to Select Measurement Points in Access Point Localization Proceeding of the International MultiConference of Engineer and Computer Scientit 205 Vol II, IMECS 205, March 8-20, 205, Hong Kong How to Select Meaurement Point in Acce Point Localization Xiaoling Yang,

More information

xy-monotone path existence queries in a rectilinear environment

xy-monotone path existence queries in a rectilinear environment CCCG 2012, Charlottetown, P.E.I., Augut 8 10, 2012 xy-monotone path exitence querie in a rectilinear environment Gregory Bint Anil Mahehwari Michiel Smid Abtract Given a planar environment coniting of

More information

Focused Video Estimation from Defocused Video Sequences

Focused Video Estimation from Defocused Video Sequences Focued Video Etimation from Defocued Video Sequence Junlan Yang a, Dan Schonfeld a and Magdi Mohamed b a Multimedia Communication Lab, ECE Dept., Univerity of Illinoi, Chicago, IL b Phyical Realization

More information

1 The secretary problem

1 The secretary problem Thi i new material: if you ee error, pleae email jtyu at tanford dot edu 1 The ecretary problem We will tart by analyzing the expected runtime of an algorithm, a you will be expected to do on your homework.

More information

Drawing Lines in 2 Dimensions

Drawing Lines in 2 Dimensions Drawing Line in 2 Dimenion Drawing a traight line (or an arc) between two end point when one i limited to dicrete pixel require a bit of thought. Conider the following line uperimpoed on a 2 dimenional

More information

Stochastic Search and Graph Techniques for MCM Path Planning Christine D. Piatko, Christopher P. Diehl, Paul McNamee, Cheryl Resch and I-Jeng Wang

Stochastic Search and Graph Techniques for MCM Path Planning Christine D. Piatko, Christopher P. Diehl, Paul McNamee, Cheryl Resch and I-Jeng Wang Stochatic Search and Graph Technique for MCM Path Planning Chritine D. Piatko, Chritopher P. Diehl, Paul McNamee, Cheryl Rech and I-Jeng Wang The John Hopkin Univerity Applied Phyic Laboratory, Laurel,

More information

An Approach to a Test Oracle for XML Query Testing

An Approach to a Test Oracle for XML Query Testing An Approach to a Tet Oracle for XML Query Teting Dae S. Kim-Park, Claudio de la Riva, Javier Tuya Univerity of Oviedo Computing Department Campu of Vieque, /n, 33204 (SPAIN) kim_park@li.uniovi.e, claudio@uniovi.e,

More information

Key Terms - MinMin, MaxMin, Sufferage, Task Scheduling, Standard Deviation, Load Balancing.

Key Terms - MinMin, MaxMin, Sufferage, Task Scheduling, Standard Deviation, Load Balancing. Volume 3, Iue 11, November 2013 ISSN: 2277 128X International Journal of Advanced Reearch in Computer Science and Software Engineering Reearch Paper Available online at: www.ijarce.com Tak Aignment in

More information

Lecture 14: Minimum Spanning Tree I

Lecture 14: Minimum Spanning Tree I COMPSCI 0: Deign and Analyi of Algorithm October 4, 07 Lecture 4: Minimum Spanning Tree I Lecturer: Rong Ge Scribe: Fred Zhang Overview Thi lecture we finih our dicuion of the hortet path problem and introduce

More information

LinkGuide: Towards a Better Collection of Hyperlinks in a Website Homepage

LinkGuide: Towards a Better Collection of Hyperlinks in a Website Homepage Proceeding of the World Congre on Engineering 2007 Vol I LinkGuide: Toward a Better Collection of Hyperlink in a Webite Homepage A. Ammari and V. Zharkova chool of Informatic, Univerity of Bradford anammari@bradford.ac.uk,

More information

Minimum congestion spanning trees in bipartite and random graphs

Minimum congestion spanning trees in bipartite and random graphs Minimum congetion panning tree in bipartite and random graph M.I. Otrovkii Department of Mathematic and Computer Science St. John Univerity 8000 Utopia Parkway Queen, NY 11439, USA e-mail: otrovm@tjohn.edu

More information

Domain-Specific Modeling for Rapid System-Wide Energy Estimation of Reconfigurable Architectures

Domain-Specific Modeling for Rapid System-Wide Energy Estimation of Reconfigurable Architectures Domain-Specific Modeling for Rapid Sytem-Wide Energy Etimation of Reconfigurable Architecture Seonil Choi 1,Ju-wookJang 2, Sumit Mohanty 1, Viktor K. Praanna 1 1 Dept. of Electrical Engg. 2 Dept. of Electronic

More information

/06/$ IEEE 364

/06/$ IEEE 364 006 IEEE International ympoium on ignal Proceing and Information Technology oie Variance Etimation In ignal Proceing David Makovoz IPAC, California Intitute of Technology, MC-0, Paadena, CA, 95 davidm@ipac.caltech.edu;

More information

A reduced reference image quality metric based on feature fusion and neural networks

A reduced reference image quality metric based on feature fusion and neural networks Univerity of Wollongong Reearch Online Faculty of Engineering and Information Science - Paper: Part A Faculty of Engineering and Information Science 2011 A reduced reference image quality metric baed on

More information

Performance Evaluation of an Advanced Local Search Evolutionary Algorithm

Performance Evaluation of an Advanced Local Search Evolutionary Algorithm Anne Auger and Nikolau Hanen Performance Evaluation of an Advanced Local Search Evolutionary Algorithm Proceeding of the IEEE Congre on Evolutionary Computation, CEC 2005 c IEEE Performance Evaluation

More information

KS3 Maths Assessment Objectives

KS3 Maths Assessment Objectives KS3 Math Aement Objective Tranition Stage 9 Ratio & Proportion Probabilit y & Statitic Appreciate the infinite nature of the et of integer, real and rational number Can interpret fraction and percentage

More information

Maneuverable Relays to Improve Energy Efficiency in Sensor Networks

Maneuverable Relays to Improve Energy Efficiency in Sensor Networks Maneuverable Relay to Improve Energy Efficiency in Senor Network Stephan Eidenbenz, Luka Kroc, Jame P. Smith CCS-5, MS M997; Lo Alamo National Laboratory; Lo Alamo, NM 87545. Email: {eidenben, kroc, jpmith}@lanl.gov

More information

PARALLEL knn ON GPU ARCHITECTURE USING OpenCL

PARALLEL knn ON GPU ARCHITECTURE USING OpenCL PARALLEL knn ON GPU ARCHITECTURE USING OpenCL V.B.Nikam 1, B.B.Mehram 2 1 Aociate Profeor, Department of Computer Engineering and Information Technology, Jijabai Technological Intitute, Matunga, Mumbai,

More information

Testing Structural Properties in Textual Data: Beyond Document Grammars

Testing Structural Properties in Textual Data: Beyond Document Grammars Teting Structural Propertie in Textual Data: Beyond Document Grammar Felix Saaki and Jen Pönninghau Univerity of Bielefeld, Germany Abtract Schema language concentrate on grammatical contraint on document

More information

Computer Arithmetic Homework Solutions. 1 An adder for graphics. 2 Partitioned adder. 3 HDL implementation of a partitioned adder

Computer Arithmetic Homework Solutions. 1 An adder for graphics. 2 Partitioned adder. 3 HDL implementation of a partitioned adder Computer Arithmetic Homework 3 2016 2017 Solution 1 An adder for graphic In a normal ripple carry addition of two poitive number, the carry i the ignal for a reult exceeding the maximum. We ue thi ignal

More information

Increasing Throughput and Reducing Delay in Wireless Sensor Networks Using Interference Alignment

Increasing Throughput and Reducing Delay in Wireless Sensor Networks Using Interference Alignment Int. J. Communication, Network and Sytem Science, 0, 5, 90-97 http://dx.doi.org/0.436/ijcn.0.50 Publihed Online February 0 (http://www.scirp.org/journal/ijcn) Increaing Throughput and Reducing Delay in

More information

Learning-Based Quality Control for Cardiac MR Images

Learning-Based Quality Control for Cardiac MR Images 1 Learning-Baed Quality Control for Cardiac MR Image Giacomo Tarroni, Ozan Oktay, Wenjia Bai, Andrea Schuh, Hideaki Suzuki, Jonathan Paerat-Palmbach, Antonio de Marvao, Declan P. O Regan, Stuart Cook,

More information

Compressed Sensing Image Processing Based on Stagewise Orthogonal Matching Pursuit

Compressed Sensing Image Processing Based on Stagewise Orthogonal Matching Pursuit Senor & randucer, Vol. 8, Iue 0, October 204, pp. 34-40 Senor & randucer 204 by IFSA Publihing, S. L. http://www.enorportal.com Compreed Sening Image Proceing Baed on Stagewie Orthogonal Matching Puruit

More information

Chapter 13 Non Sampling Errors

Chapter 13 Non Sampling Errors Chapter 13 Non Sampling Error It i a general aumption in the ampling theory that the true value of each unit in the population can be obtained and tabulated without any error. In practice, thi aumption

More information

( ) subject to m. e (2) L are 2L+1. = s SEG SEG Las Vegas 2012 Annual Meeting Page 1

( ) subject to m. e (2) L are 2L+1. = s SEG SEG Las Vegas 2012 Annual Meeting Page 1 A new imultaneou ource eparation algorithm uing frequency-divere filtering Ying Ji*, Ed Kragh, and Phil Chritie, Schlumberger Cambridge Reearch Summary We decribe a new imultaneou ource eparation algorithm

More information

An Intro to LP and the Simplex Algorithm. Primal Simplex

An Intro to LP and the Simplex Algorithm. Primal Simplex An Intro to LP and the Simplex Algorithm Primal Simplex Linear programming i contrained minimization of a linear objective over a olution pace defined by linear contraint: min cx Ax b l x u A i an m n

More information

(12) Patent Application Publication (10) Pub. No.: US 2011/ A1

(12) Patent Application Publication (10) Pub. No.: US 2011/ A1 (19) United State US 2011 0316690A1 (12) Patent Application Publication (10) Pub. No.: US 2011/0316690 A1 Siegman (43) Pub. Date: Dec. 29, 2011 (54) SYSTEMAND METHOD FOR IDENTIFYING ELECTRICAL EQUIPMENT

More information

ANALYSIS OF THE FIRST LAYER IN WEIGHTLESS NEURAL NETWORKS FOR 3_DIMENSIONAL PATTERN RECOGNITION

ANALYSIS OF THE FIRST LAYER IN WEIGHTLESS NEURAL NETWORKS FOR 3_DIMENSIONAL PATTERN RECOGNITION ANALYSIS OF THE FIRST LAYER IN WEIGHTLESS NEURAL NETWORKS FOR 3_DIMENSIONAL PATTERN RECOGNITION A. Váque-Nava * Ecuela de Ingeniería. CENTRO UNIVERSITARIO MEXICO. DIVISION DE ESTUDIOS SUPERIORES J. Figueroa

More information

Aspects of Formal and Graphical Design of a Bus System

Aspects of Formal and Graphical Design of a Bus System Apect of Formal and Graphical Deign of a Bu Sytem Tiberiu Seceleanu Univerity of Turku, Dpt. of Information Technology Turku, Finland tiberiu.eceleanu@utu.fi Tomi Weterlund Turku Centre for Computer Science

More information

Evolution of Non-Deterministic Incremental Algorithms. Hugues Juille. Volen Center for Complex Systems. Brandeis University. Waltham, MA

Evolution of Non-Deterministic Incremental Algorithms. Hugues Juille. Volen Center for Complex Systems. Brandeis University. Waltham, MA Evolution of Non-Determinitic Incremental Algorithm a a New Approach for Search in State Space Hugue Juille Computer Science Department Volen Center for Complex Sytem Brandei Univerity Waltham, MA 02254-9110

More information

Cutting Stock by Iterated Matching. Andreas Fritsch, Oliver Vornberger. University of Osnabruck. D Osnabruck.

Cutting Stock by Iterated Matching. Andreas Fritsch, Oliver Vornberger. University of Osnabruck. D Osnabruck. Cutting Stock by Iterated Matching Andrea Fritch, Oliver Vornberger Univerity of Onabruck Dept of Math/Computer Science D-4909 Onabruck andy@informatikuni-onabrueckde Abtract The combinatorial optimization

More information

Chapter S:II (continued)

Chapter S:II (continued) Chapter S:II (continued) II. Baic Search Algorithm Sytematic Search Graph Theory Baic State Space Search Depth-Firt Search Backtracking Breadth-Firt Search Uniform-Cot Search AND-OR Graph Baic Depth-Firt

More information

Laboratory Exercise 6

Laboratory Exercise 6 Laboratory Exercie 6 Adder, Subtractor, and Multiplier a a The purpoe of thi exercie i to examine arithmetic circuit that add, ubtract, and multiply number. Each b c circuit will be decribed in Verilog

More information

New Structural Decomposition Techniques for Constraint Satisfaction Problems

New Structural Decomposition Techniques for Constraint Satisfaction Problems New Structural Decompoition Technique for Contraint Satifaction Problem Yaling Zheng and Berthe Y. Choueiry Contraint Sytem Laboratory Univerity of Nebraka-Lincoln Email: yzheng choueiry@ce.unl.edu Abtract.

More information

A Fast Association Rule Algorithm Based On Bitmap and Granular Computing

A Fast Association Rule Algorithm Based On Bitmap and Granular Computing A Fat Aociation Rule Algorithm Baed On Bitmap and Granular Computing T.Y.Lin Xiaohua Hu Eric Louie Dept. of Computer Science College of Information Science IBM Almaden Reearch Center San Joe State Univerity

More information

Laboratory Exercise 6

Laboratory Exercise 6 Laboratory Exercie 6 Adder, Subtractor, and Multiplier The purpoe of thi exercie i to examine arithmetic circuit that add, ubtract, and multiply number. Each circuit will be decribed in Verilog and implemented

More information

IMPLEMENTATION OF CHORD LENGTH SAMPLING FOR TRANSPORT THROUGH A BINARY STOCHASTIC MIXTURE

IMPLEMENTATION OF CHORD LENGTH SAMPLING FOR TRANSPORT THROUGH A BINARY STOCHASTIC MIXTURE Nuclear Mathematical and Computational Science: A Century in Review, A Century Anew Gatlinburg, Tenneee, April 6-, 003, on CD-ROM, American Nuclear Society, LaGrange Park, IL (003) IMPLEMENTATION OF CHORD

More information

Quadrilaterals. Learning Objectives. Pre-Activity

Quadrilaterals. Learning Objectives. Pre-Activity Section 3.4 Pre-Activity Preparation Quadrilateral Intereting geometric hape and pattern are all around u when we tart looking for them. Examine a row of fencing or the tiling deign at the wimming pool.

More information

Tracking High Speed Skater by Using Multiple Model

Tracking High Speed Skater by Using Multiple Model Vol. 2, No. 26 Tracing High Speed Sater by Uing Multiple Model Guojun Liu & Xianglong Tang School of Computer Science & Engineering Harbin Intitute of Technology Harbin 5000, China E-mail: hitliu@hit.edu.cn

More information

(A)ATSR RE-ANALYSIS FOR CLIMATE - CLOUD CLEARING METHODOLOGY

(A)ATSR RE-ANALYSIS FOR CLIMATE - CLOUD CLEARING METHODOLOGY (A)ATSR RE-ANALYSIS FOR CLIMATE - CLOUD CLEARING METHODOLOGY Chri Old, Chri Merchant Univerity of Edinburgh, The Crew Building, Wet Main Road, Edinburgh, EH9 3JN, United Kingdom Email: cold@ed.ac.uk Email:

More information

Mirror shape recovery from image curves and intrinsic parameters: Rotationally symmetric and conic mirrors. Abstract. 2. Mirror shape recovery

Mirror shape recovery from image curves and intrinsic parameters: Rotationally symmetric and conic mirrors. Abstract. 2. Mirror shape recovery Mirror hape recovery from image curve and intrinic parameter: Rotationally ymmetric and conic mirror Nuno Gonçalve and Helder Araújo Λ Intitute of Sytem and Robotic Univerity of Coimbra Pinhal de Marroco

More information

The Association of System Performance Professionals

The Association of System Performance Professionals The Aociation of Sytem Performance Profeional The Computer Meaurement Group, commonly called CMG, i a not for profit, worldwide organization of data proceing profeional committed to the meaurement and

More information

Frequency Table Computation on Dataflow Architecture

Frequency Table Computation on Dataflow Architecture Frequency Table Computation on Dataflow Architecture P. Škoda *, V. Sruk **, and B. Medved Rogina * * Ruđer Bošković Intitute, Zagreb, Croatia ** Faculty of Electrical Engineering and Computing, Univerity

More information

Khoirul Umam 1, Agus Zainal Arifin 2 and Dini Adni Navastara 3

Khoirul Umam 1, Agus Zainal Arifin 2 and Dini Adni Navastara 3 I J C T A, 9(-A), 016, pp 763-777 International Science Pre A Novel Strategy of Differential Evolution Algorithm Croover Operator Baed on Graylevel Cluter Similarity for Automatic Multilevel Image Threholding

More information

Modeling of underwater vehicle s dynamics

Modeling of underwater vehicle s dynamics Proceeding of the 11th WEA International Conference on YTEM, Agio Nikolao, Crete Iland, Greece, July 23-25, 2007 44 Modeling of underwater vehicle dynamic ANDRZEJ ZAK Department of Radiolocation and Hydrolocation

More information

A Practical Model for Minimizing Waiting Time in a Transit Network

A Practical Model for Minimizing Waiting Time in a Transit Network A Practical Model for Minimizing Waiting Time in a Tranit Network Leila Dianat, MASc, Department of Civil Engineering, Sharif Univerity of Technology, Tehran, Iran Youef Shafahi, Ph.D. Aociate Profeor,

More information

Web Science and additionality

Web Science and additionality Admin tuff... Lecture 1: EITN01 Web Intelligence and Information Retrieval Meage, lide, handout, lab manual and link: http://www.eit.lth.e/coure/eitn01 Contact: Ander Ardö, Ander.Ardo@eit.lth.e, room:

More information

Locating Brain Tumors from MR Imagery Using Symmetry

Locating Brain Tumors from MR Imagery Using Symmetry ocating rain Tumor from M magery Uing Symmetry Nilanjan ay aidya Nath Saha and Matthew obert Graham rown {nray1 baidya mbrown}@cualbertaca epartment of Computing Science Univerity of lberta Canada btract

More information

Topics. Lecture 37: Global Optimization. Issues. A Simple Example: Copy Propagation X := 3 B > 0 Y := 0 X := 4 Y := Z + W A := 2 * 3X

Topics. Lecture 37: Global Optimization. Issues. A Simple Example: Copy Propagation X := 3 B > 0 Y := 0 X := 4 Y := Z + W A := 2 * 3X Lecture 37: Global Optimization [Adapted from note by R. Bodik and G. Necula] Topic Global optimization refer to program optimization that encompa multiple baic block in a function. (I have ued the term

More information

Introduction to PET Image Reconstruction. Tomographic Imaging. Projection Imaging. PET Image Reconstruction 11/6/07

Introduction to PET Image Reconstruction. Tomographic Imaging. Projection Imaging. PET Image Reconstruction 11/6/07 Introduction to PET Image Recontruction Adam Aleio Nuclear Medicine Lecture Imaging Reearch Laboratory Diviion of Nuclear Medicine Univerity of Wahington Fall 2007 http://dept.wahington.edu/nucmed/irl/education.html

More information

Advanced Encryption Standard and Modes of Operation

Advanced Encryption Standard and Modes of Operation Advanced Encryption Standard and Mode of Operation G. Bertoni L. Breveglieri Foundation of Cryptography - AES pp. 1 / 50 AES Advanced Encryption Standard (AES) i a ymmetric cryptographic algorithm AES

More information

AUTOMATIC TEST CASE GENERATION USING UML MODELS

AUTOMATIC TEST CASE GENERATION USING UML MODELS Volume-2, Iue-6, June-2014 AUTOMATIC TEST CASE GENERATION USING UML MODELS 1 SAGARKUMAR P. JAIN, 2 KHUSHBOO S. LALWANI, 3 NIKITA K. MAHAJAN, 4 BHAGYASHREE J. GADEKAR 1,2,3,4 Department of Computer Engineering,

More information

Kinematics Programming for Cooperating Robotic Systems

Kinematics Programming for Cooperating Robotic Systems Kinematic Programming for Cooperating Robotic Sytem Critiane P. Tonetto, Carlo R. Rocha, Henrique Sima, Altamir Dia Federal Univerity of Santa Catarina, Mechanical Engineering Department, P.O. Box 476,

More information

An Algebraic Approach to Adaptive Scalable Overlay Network Monitoring

An Algebraic Approach to Adaptive Scalable Overlay Network Monitoring An Algebraic Approach to Adaptive Scalable Overlay Network Monitoring ABSTRACT Overlay network monitoring enable ditributed Internet application to detect and recover from path outage and period of degraded

More information

Data Mining with Linguistic Thresholds

Data Mining with Linguistic Thresholds Int. J. Contemp. Math. Science, Vol. 7, 22, no. 35, 7-725 Data Mining with Linguitic Threhold Tzung-Pei Hong Department of Electrical Engineering National Univerity of Kaohiung Kaohiung, Taiwan, R.O.C.

More information

A Study of a Variable Compression Ratio and Displacement Mechanism Using Design of Experiments Methodology

A Study of a Variable Compression Ratio and Displacement Mechanism Using Design of Experiments Methodology A Study of a Variable Compreion Ratio and Diplacement Mechanim Uing Deign of Experiment Methodology Shugang Jiang, Michael H. Smith, Maanobu Takekohi Abtract Due to the ever increaing requirement for engine

More information

The LLAhclust Package

The LLAhclust Package Verion 0.2-1 Date 2007-31-08 The LLAhclut Package September 1, 2007 Title Hierarchical clutering of variable or object baed on the likelihood linkage analyi method Author Ivan Kojadinovic, Iraël-Céar Lerman,

More information