La quantité d'images disponible sur Internet ne fait que croître, engendrant un besoin d'algorithmes permettant de fouiller ces images et retrouver de l'information. Les systèmes de recherche d'images par le contenu ont été développées dans ce but. Mais les bases de données grandissant, de nouveaux défis sont apparus. Dans cette thèse, la classification à grain fin est étudiée en particulier. Elle consiste à séparer des images qui sont relativement semblables visuellement mais représentent différents concepts, et à regrouper des images qui sont différentes visuellement mais représentent le même concept. Il est montré dans un premier temps que les techniques classiques de recherche d'images par le contenu rencontrent des difficultés à effectuer cette tâche. Même les techniques utilisant les machines à vecteur de support (SVM), qui sont très performants pour la classification, n'y parviennent pas complètement. Ces techniques n'explorent souvent pas assez l'espace de recherche pour résoudre ce problème. D'autres méthodes, comme les algorithmes évolutionnaires sont également étudiées pour leur capacité à identifier des zones intéressantes de l'espace de recherche en un temps raisonnable. Toutefois, leurs performances restent encore limitées. Par conséquent, l'apport de la thèse consiste à proposer un système hybride combinant un algorithme évolutionnaire et un SVM a finalement été développé. L'algorithme évolutionnaire est utilisé pour construire itérativement un ensemble d'apprentissage pour le SVM. Ce système est évalué avec succès sur la base de données Caltech-256 contenant envieront 30000 images réparties en 256 catégories / Given the ever growing amount of visual content available on the Internet, the need for systems able to search through this content has grown. Content based image retrieval systems have been developed to address this need. But with the growing size of the databases, new challenges arise. In this thesis, the fine grained classification problem is studied in particular. It is first shown that existing techniques, and in particular the support vector machines which are one of the best image classification technique, have some difficulties in solving this problem. They often lack of exploration in their process. Then, evolutionary algorithms are considered to solve the problem, for their balance between exploration and exploitation. But their performances are not good enough either. Finally, an hybrid system combining an evolutionary algorithm and a support vector machine is proposed. This system uses the evolutionary algorithm to iteratively feed the support vector machine with training samples. The experiments conducted on Caltech-256, a state of the art database containing around 30000 images, show very encouraging results

Object categorization is one of the fundamental topics in computer vision research. Most current work in object categorization aims to discriminate among generic object classes with gross differences. However, many applications require much finer distinctions. This thesis focuses on the design, evaluation and analysis of learning algorithms for fine- grained object classification. The contributions of the thesis are three-fold. First, we introduce two databases of high-resolution images of arthropod specimens we collected to promote the development of highly accurate fine-grained recognition methods. Second, we give a literature review on the development of Bag-of-words (BOW) approaches to image classification and present the stacked evidence tree approach we developed for the fine-grained classification task. We draw connections and analyze differences between those two genres of approaches, which leads to a better understanding about the design of image classification approaches. Third, benchmark results on our two datasets are pre- sented. We further analyze the influence of two important variables on the performance of fine-grained classification. The experiments corroborate our hypotheses that a) high resolution images and b) more aggressive information extraction, such as finer descriptor encoding with large dictionaries or classifiers based on raw descriptors, is required to achieve good fine-grained categorization accuracy. / Graduation date: 2013

In this study, to increase inference speed of convolutional neural networks, the optimization technique low-rank tensor decomposition has been implemented and applied to AlexNet which had been trained to classify dog breeds. Due to a small training set, transfer learning was used in order to be able to classify dog breeds. The purpose of the study is to investigate how effective low-rank tensor decomposition is when the training set is limited. The results obtained from this study, compared to a previous study, indicate that there is a strong relationship between the effects of the tensor decomposition and how much available training data exists. A significant speed up can be obtained in the different convolutional layers using tensor decomposition. However, since there is a need to retrain the network after the decomposition and due to the limited dataset there is a slight decrease in accuracy. / För att öka inferenshastigheten hos faltningssnätverk, har i denna studie optimeringstekniken low-rank tensor decomposition implementerats och applicerats på AlexNet, som har tränats för att klassificera hundraser. På grund av en begränsad mängd träningsdata användes transfer learning för uppgiften. Syftet med studien är att undersöka hur effektiv low-rank tensor decomposition är när träningsdatan är begränsad. Jämfört med resultaten från en tidigare studie visar resultaten från denna studie att det finns ett starkt samband mellan effekterna av low-rank tensor decomposition och hur mycket tillgänglig träningsdata som finns. En signifikant hastighetsökning kan uppnås i de olika faltningslagren med hjälp av low-rank tensor decomposition. Eftersom det finns ett behov av att träna om nätverket efter dekompositionen och på grund av den begränsade mängden data så uppnås hastighetsökningen dock på bekostnad av en viss minskning i precisionen för modellen.