| Face recognition has been studied extensively from past four decades and still an active research area in the field of image processing and machine vision due to the fundamental challenges it possesses and potential applications where human identification or verification is needed.Face recognition system utilizes low cost equipment and more importantly the person under investigation doesn't need any direct contact during image acquisition are reckoned the most important features of face recognition over other biometric investigation methods.Even thought,the history of research on face recognition is quite long and lot of progress has been made in recent years,the performance of existing face recognition systems has yet not reached at the level of satisfaction under the challenging conditions.Therefore,face recognition is considered a particular and hard case of pattern or object recognition.With the wide spread use of smartphones and fast mobile networks,millions of photos are uploaded every day to the cloud storages such as Dropbox or social networks such as Facebook,Twitter,Instagram,Google+,and Flicker.Organizing and retrieving relevant information from these photos is very challenging and directly impact user experience on those platforms.For example,users commonly look for photos that were taken at a particular location,at a particular time,or with a particular friend.The former two queries are fairly straightforward,as almost all of today's cameras embed time and GPS location into photos.The last query,i.e.contextual query,is more challenging as there is no explicit signal about the identities of people in the photos.The key for this identification is the detection of human faces.This has made low complexity,rapid and accurate face detection an essential component for cloud based photo sharing/storage platforms.The research field of face recognition has quite long history and lot of techniques have been presented by different researchers on this topic.In order to systematically represent the background of this research field,face recognition system has to be categorized with respect to the different approaches used for face recognition.In general,two distinct approaches namely feature based approach and holistic approach are used for face recognition.Feature based approaches use distinctive facial features such as nose,eyes,mouth etc.and compute geometric relationships between different facial points.The early attempts in this research area were made in 1973.Kanade used simple image processing techniques to extract the feature vector by focusing on 16 distinct face features.A simple database of 20 people was used and parameters such as area,distance or angles ratios are calculated between two images of same candidate.In this early attempt by Kanade,a recognition rate of 75%was achieved which was considered quite remarkable at that time.Later on,more progress is made in feature extraction based approaches and new techniques have been proposed.Some of the popular feature extraction based techniques are deformable templates approach,Hough transform methods,mathematical morphological operators etc.However,these techniques better work in constrained conditions and in real situation,their performance also declines.One of the widely used feature based technique which has successfully represented the facial features in particular databases,namely elastic bench graph matching was proposed by Wiskott et al.which works on the basis of dynamic link structure.Using this technique,a recognition rate of 98%is achieved while using a database of 250 people.Finally,statistical features extraction approach,which extracts the textural features of image gain popularity due to its invariance to monotonic gray level changes and computational efficiency.The local binary pattern,which was originally designed for texture analysis was proposed by Ojala et al.(1996)is a 3x3 local descriptor which thresholds the neighboring pixels to its central pixel and compute binary pattern.After Ojala's presentation of LBP descriptor,plenty of research work is done on this texture based descriptor and currently LBP has lot of extended shapes.Thus,the idea for face description by using LBP is motivated by the fact that faces are composed of micro-patterns and these micro-patterns are well defined by this method.The Holistic approach try to recognize faces using entire image rather than local features.The most popular approach used for face recognition is principal component analysis first proposed by Turk and Pentland,1991.It computes the Eigen faces of the co-variance matrices.PCA is the most widely used algorithm which not only used as a base to compare new methods until today but also many new algorithms are derived from this method.Many hybrid methods are developed for face recognition and most of them utilized PCA to reduce the dimensionality of feature vector.Some of the modern software has utilized artificial intelligence to train the neural networks for face recognition.DeepFace software used by Facebook is one the best example of modern face recognition system that uses neural networks.Face recognition has always been an active research area in the vision community.It is a challenging problem in the field of image analysis and computer vision,and as such has received a great deal of attention over the last few years because of its many applications in various domains.Plenty of research has been done in this research field and under controlled conditions,fruitful results have been achieved.But under uncontrolled conditions,the face recognition is still an unsolved problem.For example one key application of face recognition is security surveillance where images can't be collected under controlled conditions.Therefore,a texture based statistical feature extraction method namely Local binary pattern is proposed in this research to cope with some issues of face recognition.The texture of human face contains a lot of information in order to design an identification system for human face.LBP is a simple but a powerful tool to extract textural features.The face area is first divided into small regions from which Local Binary Pattern(LBP)histograms are extracted and concatenated into a single,feature histogram efficiently representing the face image.The recognition is performed using a Chi square distance method as a dissimilarity measure.The main objective of this research is to achieve a comparable recognition rate with previously experimented techniques and to speed up the recognition process by reducing the length of feature vector.This thesis identifies the general difficulties that influence the rate of face recognition and tries to improve the performance of recognition rates in particular to the variations in facial expressions and intensity variation effects on face.The local binary pattern which extracts textural features from the surface of the images is used for experimentation.A wide verity of LBP schemes is tested.The experiments are carried out by varying the size of parameters of LBP and results are compared with other state of the art face recognition methods.Standard FERET database which contains different sets of images with varying pose,illumination,expressions etc.and which are used widely for face recognition research has been used for experimentation.Few extensions of LBP are presented after initial experiments.These extensions include the expansion of neighborhood around examined pixel(multi-circle neighborhood),the weight assignment procedure and use of PCA with LBP.These extensions are implemented practically and results are compared with previously experimented schemes of LBP.Favorable results were produced especially using the principal component analysis with Local binary pattern.The most important Eigen faces were selected from feature vector of training images and comparison was made.The result shows that recognition rate with and without applying PCA was almost same but the length of feature vector reduced considerably with use of PCA.In short,it can be concluded that with the proper selection of correct scheme of LBP,best results can be achieved for face recognition.The resulting rank curves shows that LBP is one to the best method for feature based face recognition system.With fb Set of FERET database,a recognition rate of about 97 percent is achieved which is quite remarkable.As fb Set contained the images with varying facial expressions,thus,it can be concluded that LBP is quite robust for varying expressions.On the other hand,although the recognition rate with other data sets were quite poor but still better than other approaches used for same data sets. |
PDF Full Text Request