Object recognition is a hugely researched domain that employs methods derived from mathematics, physics and biology. This thesis combines the approaches for object classification that base on two features – color and shape. Color is represented by color histograms and shape by skeletal graphs. Four hybrids are proposed which combine those approaches in different manners and the hybrids are then tested to find out which of them gives best results. / Mail the author at radoslaw.cichocki(at)gmail.com

Within this thesis an algorithm for object recognition called Cluster Matching has been developed, implemented and evaluated. The image information is sampled at arbitrary sample points, instead of interest points, and local image features are extracted. These sample points are used as a compact representation of the image data and can quickly be searched for prior known objects. The algorithm is evaluated on a test set of images and the result is surprisingly reliable and time efficient.

The vivid memory of an emotional event, as well as memory for incidental details associated with the arousing event, has been referred to collectively as a “flashbulb memory”. An important aspect of flashbulb memory in people is that an emotional event enhances memory of contextual details, such as the weather, or clothes one was wearing at the time of the event. Therefore, an emotional event not only produces a detailed memory of the event, itself, but also enhances memory for contextual details that would otherwise not be remembered. The first goal of this work is to describe the development of my animal model of flashbulb memory, including a discussion of the importance of the timing between an emotional event and incidental, contextual cues. The second goal is to address the time-dependent neuroendocrine processes involved in stress-induced memory enhancement in rats. The involvement of brain structures, namely the hippocampus and amygdala, and hormones, including corticosterone and epinephrine, that interact to produce a composite memory of the contextual cues occurring in close temporal proximity to an emotional event are discussed. The results of Experiment 1 validate the animal model of flashbulb memory whereby an emotional event (predator exposure) produced memory for context cues that, under control conditions, would be forgotten. This memory enhancement only occurred when the emotional event was close in temporal proximity to training in the task. Experiment 2 provided evidence that epinephrine administration close in time to training mimicked the context memory formation induced by brief predator exposure, while propranolol, a β-adrenergic antagonist, as well as CPP, an NMDA receptor antagonist, blocked this effect. The results of Experiment 3 revealed that propranolol, CPP, and dexamethasone also blocked the brief predator stress-induced context memory formation. The results of Experiment 4 revealed that cannulated animals infused with aCSF (control) did not show evidence of predator stress-induced memory, therefore methodological issues within this experiment are addressed. Finally, the results of Experiment 5 revealed that adrenalectomy eliminated the predator stress-induced context memory compared to sham operated animals, suggesting that endogenous stress hormones are required for stress-induced context memory formation. Further, adrenalectomized rats supplemented with epinephrine before training did show evidence of context memory enhancement suggesting that epinephrine eliminated the memory impairment produced by adrenalectomy, and was sufficient to enhance memory in the absence of corticosterone. Overall this approach has provided insight into the time-dependent neuroendocrine processes involved in the formation of flashbulb, and potentially traumatic, memories in people.

This thesis aims to understand the learning mechanisms which underpin the process of visual object recognition in the primate ventral visual system. The computational crux of this problem lies in the ability to retain specificity to recognize particular objects or faces, while exhibiting generality across natural variations and distortions in the view (DiCarlo et al., 2012). In particular, the work presented is focussed on gaining insight into the processes through which transformation-invariant visual representations may develop in the primate ventral visual system. The primary motivation for this work is the belief that some of the fundamental mechanisms employed in the primate visual system may only be captured through modelling the individual action potentials of neurons and therefore, existing rate-coded models of this process constitute an inadequate level of description to fully understand the learning processes of visual object recognition. To this end, spiking neural network models are formulated and applied to the problem of learning transformation-invariant visual representations, using a spike-time dependent learning rule to adjust the synaptic efficacies between the neurons. The ways in which the existing rate-coded CT (Stringer et al., 2006) and Trace (Földiák, 1991) learning mechanisms may operate in a simple spiking neural network model are explored, and these findings are then applied to a more accurate model using realistic 3-D stimuli. Three mechanisms are then examined, through which a spiking neural network may solve the problem of learning separate transformation-invariant representations in scenes composed of multiple stimuli by temporally segmenting competing input representations. The spike-time dependent plasticity in the feed-forward connections is then shown to be able to exploit these input layer dynamics to form individual stimulus representations in the output layer. Finally, the work is evaluated and future directions of investigation are proposed.

L'objectif de cette thèse est de concevoir un système automatique constitué d'une ou plusieurs caméras capables de détecter en trois dimensions un amalgame d'objets stockés dans un conteneur. Pour ceci, il est nécessaire de modéliser, de reconnaître et de localiser des formes dans une image. Dans un premier temps, Nous proposons une solution d'optimisation du calibrage de caméras. C'est une tâche essentielle pour récupérer des informations quantitatives sur les images capturées. Cette méthode nécessite des compétences spécifiques en matière de traitement d'image, ce qui n'est pas toujours le cas dans l'industrie. Nous proposons d'automatiser et d'optimiser le système d'étalonnage en éliminant la sélection des images par l'opérateur. Ensuite, nous proposons d'améliorer les systèmes de détection d'objets fins et sans motif. Enfin, nous proposons d'adapter des algorithmes évolutionnaires dans le but d'optimiser les temps de recherche. / The aim of this thesis is to design an automatic system involving one or several cameras capable of detecting in three dimensions a set of abjects placed in a bin. To do this, we must model, recognize and locate shapes in an image. First, we propose a solution to optimize the camera calibration system. This is an essential task for the retrieval of quantitative information about the captured images. However, the current methods require specific skills in image processing, which are not always available in industry. We propose to automate and optimize the calibration system by eliminating the selection of images by the operator. Second, we propose to improve the detection systems for thin and featureless abjects. Finally, we propose to adapt evolutionary algorithms to optimize search times.

Abstract Automatic three-dimensional scene reconstruction from multiple images is a central problem in geometric computer vision. This thesis considers topics that are related to this problem area. New models and methods are presented for various tasks in such specific domains as camera calibration, image-based modeling and image matching. In particular, the main themes of the thesis are geometric camera calibration and quasi-dense image matching. In addition, a topic related to the estimation of two-view geometric relations is studied, namely, the computation of a planar homography from corresponding conics. Further, as an example of a reconstruction system, a structure-from-motion approach is presented for modeling sewer pipes from video sequences. In geometric camera calibration, the thesis concentrates on central cameras. A generic camera model and a plane-based camera calibration method are presented. The experiments with various real cameras show that the proposed calibration approach is applicable for conventional perspective cameras as well as for many omnidirectional cameras, such as fish-eye lens cameras. In addition, a method is presented for the self-calibration of radially symmetric central cameras from two-view point correspondences. In image matching, the thesis proposes a method for obtaining quasi-dense pixel matches between two wide baseline images. The method extends the match propagation algorithm to the wide baseline setting by using an affine model for the local geometric transformations between the images. Further, two adaptive propagation strategies are presented, where local texture properties are used for adjusting the local transformation estimates during the propagation. These extensions make the quasi-dense approach applicable for both rigid and non-rigid wide baseline matching. In this thesis, quasi-dense matching is additionally applied for piecewise image registration problems which are encountered in specific object recognition and motion segmentation. The proposed object recognition approach is based on grouping the quasi-dense matches between the model and test images into geometrically consistent groups, which are supposed to represent individual objects, whereafter the number and quality of grouped matches are used as recognition criteria. Finally, the proposed approach for dense two-view motion segmentation is built on a layer-based segmentation framework which utilizes grouped quasi-dense matches for initializing the motion layers, and is applicable under wide baseline conditions.

Abstract This thesis presents a multiscale framework for the construction of affine invariant pattern recognition and registration methods. The idea in the introduced approach is to extend the given pattern to a set of affine covariant versions, each carrying slightly different information, and then to apply known affine invariants to each of them separately. The key part of the framework is the construction of the affine covariant set, and this is done by combining several scaled representations of the original pattern. The advantages compared to previous approaches include the possibility of many variations and the inclusion of spatial information on the patterns in the features. The application of the multiscale framework is demonstrated by constructing several new affine invariant methods using different preprocessing techniques, combination schemes, and final recognition and registration approaches. The techniques introduced are briefly described from the perspective of the multiscale framework, and further treatment and properties are presented in the corresponding original publications. The theoretical discussion is supported by several experiments where the new methods are compared to existing approaches. In this thesis the patterns are assumed to be gray scale images, since this is the main application where affine relations arise. Nevertheless, multiscale methods can also be applied to other kinds of patterns where an affine relation is present. An additional application of one multiscale based technique in convexity measurements is introduced. The method, called multiscale autoconvolution, can be used to build a convexity measure which is a descriptor of object shape. The proposed measure has two special features compared to existing approaches. It can be applied directly to gray scale images approximating binary objects, and it can be easily modified to produce a number of measures. The new measure is shown to be straightforward to evaluate for a given shape, and it performs well in the applications, as demonstrated by the experiments in the original paper.

In this thesis, contributions are presented in the areas of shape parsing for view-based object recognition and spatio-temporal feature description for action recognition. A probabilistic model for parsing shapes into several distinguishable parts for accurate shape recognition is presented. This approach is based on robust geometric features that permit high recognition accuracy. As the second contribution in this thesis, a binary spatio-temporal feature descriptor is presented. Recent work shows that binary spatial feature descriptors are effective for increasing the efficiency of object recognition, while retaining comparable performance to state of the art descriptors. An extension of these approaches to action recognition is presented, facilitating huge gains in efficiency due to the computational advantage of computing a bag-of-words representation with the Hamming distance. A scene's motion and appearance is encoded with a short binary string. Exploiting the binary makeup of this descriptor greatly increases the efficiency while retaining competitive recognition performance.

Computer and machine vision applications are used in numerous fields to analyze static and dynamic imagery in order to assist or automate some form of decision-making process. Advancements in sensor technologies now make it possible to capture and visualize imagery at various wavelengths (or bands) of the electromagnetic spectrum. Multispectral imaging has countless applications in various field including (but not limited to) security, defense, space, medical, manufacturing and archeology. The development of advanced algorithms to process and extract salient information from the imagery is a critical component of the overall system performance. The fundamental objectives of this research project were to investigate the benefits of combining imagery from the visual and thermal bands of the electromagnetic spectrum to improve the recognition rates and accuracy of commonly found objects in an office setting. The goal was not to find a new way to “fuse” the visual and thermal images together but rather establish a methodology to extract multispectral descriptors in order to improve a machine vision system’s ability to recognize specific classes of objects.A multispectral dataset (visual and thermal) was captured and features from the visual and thermal images were extracted and used to train support vector machine (SVM) classifiers. The SVM’s class prediction ability was evaluated separately on the visual, thermal and multispectral testing datasets. Commonly used performance metrics were applied to assess the sensitivity, specificity and accuracy of each classifier. The research demonstrated that the highest recognition rate was achieved by an expert system (multiple classifiers) that combined the expertise of the visual-only classifier, the thermal-only classifier and the combined visual-thermal classifier.

This thesis addresses the generation of behaviourally useful, robust representations of the sensory world in the context of machine vision and behaviour. The goals of the work presented in this thesis are to investigate strategies for representing the visual world in a way which is behaviourally useful, to investigate the use of a neurally inspired early perceptual organisation system upon high-level processing in an object recognition system and to investigate the use of a perceptual organisation system on driving an object-based selection process. To address these problems, a biologically inspired framework for machine attention has been developed at a high level of neural abstraction, which has been heavily inspired by the psychological and physiological literature. The framework is described in this thesis, and three system implementations, which investigate the above issues, are described and analysed in detail. The primate brain has access to a coherent representation of the external world, which appears as objects at different spatial locations. It is through these representations that appropriate behavioural responses may be generated. For example, we do not become confused by cluttered scenes or by occluded objects. The representation of the visual scene is generated in a hierarchical computing structure in the primate brain: while shape and position information are able to drive attentional selection rapidly, high-level processes such as object recognition must be performed serially, passing through an attentional bottleneck. Through the process of attentional selection, the primate visual system identifies behaviourally relevant regions of the visual scene, which allows it to prioritise serial attentional shifts towards certain locations. In primates, the process of attentional selection is complex, operating upon surface representations which are robust to occlusion. Attention itself suppresses neural activity related to distractor objects, while sustaining activity relating to the target, allowing the target object to have a clear neural representation upon which the recognition process can operate. This thesis concludes that dynamic representations that are both early and robust against occlusion have the potential to be highly useful in machine vision and behaviour applications.