The human visual system must quickly and accurately deploy task-and-object-specific processing to successfully navigate the environment, which suggests several interesting research questions: What is the nature of these strategies? Are they flexible? To what extent is this behaviour optimal given the natural statistics of the environment? In this thesis, I explored these questions using human faces, a complex and dynamic source of socially relevant information that we encounter throughout our lives. Specifically, I conducted several experiments examining the role of horizontally-oriented spatial frequency components in face identification. In Chapter 2, I use computational modelling to demonstrate that the structure conveyed by these components is maximally diagnostic for face identity, and show that selective processing of this structure predicts both face identification performance and the face inversion effect. In Chapter 3, I quantify the bandwidth utilized by human observers and relate this sampling strategy to the information structure of face stimuli. In Chapter 4, I show that the selective sampling described in Chapters 2 and 3 is driven by information from the eyes. Finally, in Chapter 5, I show that the impaired horizontal selectivity associated with face inversion is enhanced by practice identifying inverted faces. Together, these experiments characterize a stimulus with differentially diagnostic information sources that, through experience, becomes selectively processed in a manner associated with task performance. These results contribute to our understanding of expert object processing and may have implications for observers experiencing face perception deficits. / Thesis / Doctor of Philosophy (PhD)

Eyewitness recognition is acknowledged to be prone to error but there is less understanding of difficulty in discriminating unfamiliar faces. This thesis examined the effects of face perception on identification of facial composites, and on unfamiliar face image comparison. Facial composites depict face memories by reconstructing features and configurations to form a likeness. They are generally reconstructed from an unfamiliar face memory, and will be unavoidably flawed. Identification will require perception of any accurate features, by someone who is familiar with the suspect and performance is typically poor. In typical face perception, face images are processed efficiently as complete units of information. Chapter 2 explored the possibility that holistic processing of inaccurate composite configurations will impair identification of individual features. Composites were split below the eyes and misaligned to impair holistic analysis (cf. Young, Hellawell, & Jay, 1987); identification was significantly enhanced, indicating that perceptual expertise with inaccurate configurations exerts powerful effects that can be reduced by enabling featural analysis. Facial composite recognition is difficult, which means that perception and judgement will be influence by an affective recognition bias: smiles enhance perceived familiarity, while negative expressions produce the opposite effect. In applied use, facial composites are generally produced from unpleasant memories and will convey negative expression; affective bias will, therefore, be important for facial composite recognition. Chapter 3 explored the effect of positive expression on composite identification: composite expressions were enhanced, and positive affect significantly increased identification. Affective quality rather than expression strength mediated the effect, with subtle manipulations being very effective. Facial image comparison (FIC) involves discrimination of two or more face images. Accuracy in unfamiliar face matching is typically in the region of 70%, and as discrimination is difficult, may be influenced by affective bias. Chapter 4 explored the smiling face effect in unfamiliar face matching. When multiple items were compared, positive affect did not enhance performance and false positive identification increased. With a delayed matching procedure, identification was not enhanced but in contrast to face recognition and simultaneous matching, positive affect improved rejection of foil images. Distinctive faces are easier to discriminate. Chapter 5 evaluated a systematic caricature transformation as a means to increase distinctiveness and enhance discrimination of unfamiliar faces. Identification of matching face images did not improve, but successful rejection of non-matching items was significantly enhanced. Chapter 6 used face matching to explore the basis of own race bias in face perception. Other race faces were manipulated to show own race facial variation, and own race faces to show African American facial variation. When multiple face images were matched simultaneously, the transformation impaired performance for all of the images; but when images were individually matched, the transformation improved perception of other race faces and discrimination of own race faces declined. Transformation of Japanese faces to show own race dimensions produced the same pattern of effects but failed to reach significance. The results provide support for both perceptual expertise and featural processing theories of own race bias. Results are interpreted with reference to face perception theories; implications for application and future study are discussed.

This work describes ways to use a person identification based on faces on mobile devices with Android platform. A reader is introduced into a structure of this system and a way to create applications for it. Besides, there are also methods usable to the face identification. Some of these methods (used in an implementation) are described in more detail. This work also contains a description of model AAM (Active Appearance Model) for implementing in mobile devices and evaluation of used algorithms results.

Human face recognition is an important area in the field of biometrics. It has been an active area of research for several decades, but still remains a challenging problem because of the complexity of the human face. In this thesis we describe fully automatic solutions that can locate faces and then perform identification and verification. We present a solution for face localisation using eye locations. We derive an efficient representation for the decision hyperplane of linear and nonlinear Support Vector Machines (SVMs). For this we introduce the novel concept of $\rho$ and $\eta$ prototypes. The standard formulation for the decision hyperplane is reformulated and expressed in terms of the two prototypes. Different kernels are treated separately to achieve further classification efficiency and to facilitate its adaptation to operate with the fast Fourier transform to achieve fast eye detection. Using the eye locations, we extract and normalise the face for size and in-plane rotations. Our method produces a more efficient representation of the SVM decision hyperplane than the well-known reduced set methods. As a result, our eye detection subsystem is faster and more accurate. The use of fractals and fractal image coding for object recognition has been proposed and used by others. Fractal codes have been used as features for recognition, but we need to take into account the distance between codes, and to ensure the continuity of the parameters of the code. We use a method based on fractal image coding for recognition, which we call the Fractal Neighbour Distance (FND). The FND relies on the Euclidean metric and the uniqueness of the attractor of a fractal code. An advantage of using the FND over fractal codes as features is that we do not have to worry about the uniqueness of, and distance between, codes. We only require the uniqueness of the attractor, which is already an implied property of a properly generated fractal code. Similar methods to the FND have been proposed by others, but what distinguishes our work from the rest is that we investigate the FND in greater detail and use our findings to improve the recognition rate. Our investigations reveal that the FND has some inherent invariance to translation, scale, rotation and changes to illumination. These invariances are image dependent and are affected by fractal encoding parameters. The parameters that have the greatest effect on recognition accuracy are the contrast scaling factor, luminance shift factor and the type of range block partitioning. The contrast scaling factor affect the convergence and eventual convergence rate of a fractal decoding process. We propose a novel method of controlling the convergence rate by altering the contrast scaling factor in a controlled manner, which has not been possible before. This helped us improve the recognition rate because under certain conditions better results are achievable from using a slower rate of convergence. We also investigate the effects of varying the luminance shift factor, and examine three different types of range block partitioning schemes. They are Quad-tree, HV and uniform partitioning. We performed experiments using various face datasets, and the results show that our method indeed performs better than many accepted methods such as eigenfaces. The experiments also show that the FND based classifier increases the separation between classes. The standard FND is further improved by incorporating the use of localised weights. A local search algorithm is introduced to find a best matching local feature using this locally weighted FND. The scores from a set of these locally weighted FND operations are then combined to obtain a global score, which is used as a measure of the similarity between two face images. Each local FND operation possesses the distortion invariant properties described above. Combined with the search procedure, the method has the potential to be invariant to a larger class of non-linear distortions. We also present a set of locally weighted FNDs that concentrate around the upper part of the face encompassing the eyes and nose. This design was motivated by the fact that the region around the eyes has more information for discrimination. Better performance is achieved by using different sets of weights for identification and verification. For facial verification, performance is further improved by using normalised scores and client specific thresholding. In this case, our results are competitive with current state-of-the-art methods, and in some cases outperform all those to which they were compared. For facial identification, under some conditions the weighted FND performs better than the standard FND. However, the weighted FND still has its short comings when some datasets are used, where its performance is not much better than the standard FND. To alleviate this problem we introduce a voting scheme that operates with normalised versions of the weighted FND. Although there are no improvements at lower matching ranks using this method, there are significant improvements for larger matching ranks. Our methods offer advantages over some well-accepted approaches such as eigenfaces, neural networks and those that use statistical learning theory. Some of the advantages are: new faces can be enrolled without re-training involving the whole database; faces can be removed from the database without the need for re-training; there are inherent invariances to face distortions; it is relatively simple to implement; and it is not model-based so there are no model parameters that need to be tweaked.

Human face recognition is an important area in the field of biometrics. It has been an active area of research for several decades, but still remains a challenging problem because of the complexity of the human face. In this thesis we describe fully automatic solutions that can locate faces and then perform identification and verification. We present a solution for face localisation using eye locations. We derive an efficient representation for the decision hyperplane of linear and nonlinear Support Vector Machines (SVMs). For this we introduce the novel concept of $\rho$ and $\eta$ prototypes. The standard formulation for the decision hyperplane is reformulated and expressed in terms of the two prototypes. Different kernels are treated separately to achieve further classification efficiency and to facilitate its adaptation to operate with the fast Fourier transform to achieve fast eye detection. Using the eye locations, we extract and normalise the face for size and in-plane rotations. Our method produces a more efficient representation of the SVM decision hyperplane than the well-known reduced set methods. As a result, our eye detection subsystem is faster and more accurate. The use of fractals and fractal image coding for object recognition has been proposed and used by others. Fractal codes have been used as features for recognition, but we need to take into account the distance between codes, and to ensure the continuity of the parameters of the code. We use a method based on fractal image coding for recognition, which we call the Fractal Neighbour Distance (FND). The FND relies on the Euclidean metric and the uniqueness of the attractor of a fractal code. An advantage of using the FND over fractal codes as features is that we do not have to worry about the uniqueness of, and distance between, codes. We only require the uniqueness of the attractor, which is already an implied property of a properly generated fractal code. Similar methods to the FND have been proposed by others, but what distinguishes our work from the rest is that we investigate the FND in greater detail and use our findings to improve the recognition rate. Our investigations reveal that the FND has some inherent invariance to translation, scale, rotation and changes to illumination. These invariances are image dependent and are affected by fractal encoding parameters. The parameters that have the greatest effect on recognition accuracy are the contrast scaling factor, luminance shift factor and the type of range block partitioning. The contrast scaling factor affect the convergence and eventual convergence rate of a fractal decoding process. We propose a novel method of controlling the convergence rate by altering the contrast scaling factor in a controlled manner, which has not been possible before. This helped us improve the recognition rate because under certain conditions better results are achievable from using a slower rate of convergence. We also investigate the effects of varying the luminance shift factor, and examine three different types of range block partitioning schemes. They are Quad-tree, HV and uniform partitioning. We performed experiments using various face datasets, and the results show that our method indeed performs better than many accepted methods such as eigenfaces. The experiments also show that the FND based classifier increases the separation between classes. The standard FND is further improved by incorporating the use of localised weights. A local search algorithm is introduced to find a best matching local feature using this locally weighted FND. The scores from a set of these locally weighted FND operations are then combined to obtain a global score, which is used as a measure of the similarity between two face images. Each local FND operation possesses the distortion invariant properties described above. Combined with the search procedure, the method has the potential to be invariant to a larger class of non-linear distortions. We also present a set of locally weighted FNDs that concentrate around the upper part of the face encompassing the eyes and nose. This design was motivated by the fact that the region around the eyes has more information for discrimination. Better performance is achieved by using different sets of weights for identification and verification. For facial verification, performance is further improved by using normalised scores and client specific thresholding. In this case, our results are competitive with current state-of-the-art methods, and in some cases outperform all those to which they were compared. For facial identification, under some conditions the weighted FND performs better than the standard FND. However, the weighted FND still has its short comings when some datasets are used, where its performance is not much better than the standard FND. To alleviate this problem we introduce a voting scheme that operates with normalised versions of the weighted FND. Although there are no improvements at lower matching ranks using this method, there are significant improvements for larger matching ranks. Our methods offer advantages over some well-accepted approaches such as eigenfaces, neural networks and those that use statistical learning theory. Some of the advantages are: new faces can be enrolled without re-training involving the whole database; faces can be removed from the database without the need for re-training; there are inherent invariances to face distortions; it is relatively simple to implement; and it is not model-based so there are no model parameters that need to be tweaked.

Face recognition finds various applications in surveillance, Law enforcement etc. These applications require fast image processing in real time. Modern GPUs have evolved fully programmable parallel stream processors. The problem of face recognition in real time system is benefited by parallelism. With the aim of fulfilling both speed and accuracy criteria we present a GPU accelerated Face Recognition system. OpenCL is a heterogeneous computing language that allows extracting parallelism on different platforms like DSP processors, FPGAs, GPUs. The proposed kernel on GPU exploits coarse grain parallelism for Local Binary Pattern (LBP) histogram computation and ELTP (Enhanced Local Ternary Pattern) feature extraction. The proposed optimization techniques on local memory, work group size and work group dimension enhances the computation of face recognition on GPU further. As a result, we have achieved a speed up of 30 times to 300 times for 124*124 to 2048*2048 image sizes for LBP and ELTP feature extraction compared to CPU. We also present a robust real time face recognition and tracking on GPU using fusion of RGB and Depth images taken from Kinect sensor. The proposed segmentation after detection algorithm enhances the performances of recognition using LBP.