In this thesis, we first present a contribution to overcome this problem of robustness for the recognition of object instances, then we straightly extend this contribution to the detection and localization of classes of objects. In a first step, we have developed a method inspired by graph matching to address the problem of fast recognition of instances of specific objects in noisy conditions. This method allows to easily combine any types of local features (eg contours, textures ...) less affected by noise than keypoints, while bypassing the normalization problem and without penalizing too much the detection speed. Unlike other methods based on a global rigid transformation, our approach is robust to complex deformations such as those due to perspective or those non-rigid inherent to the model itself (e.g. a face, a flexible magazine). Our experiments on several datasets have showed the relevance of our approach. It is overall slightly less robust to occlusion than existing approaches, but it produces better performances in noisy conditions. In a second step, we have developed an approach for detecting classes of objects in the same spirit as the bag-of-visual-words model. For this we use our cascaded micro-classifiers to recognize visual words more distinctive than the classical words simply based on visual dictionaries. Training is divided into two parts: First, we generate cascades of micro-classifiers for recognizing local parts of the model pictures and then in a second step, we use a classifier to model the decision boundary between images of class and those of non-class. We show that the association of classical visual words (from keypoints patches) and our disctinctive words results in a significant improvement. The computation time is generally quite low, given the structure of the cascades that minimizes the detection time and the form of the classifier is extremely fast to evaluate.

Dans cette thèse, nous présentons tout d'abord une contribution visant à pallier ce problème de robustesse pour la reconnaissance d'instances, puis une extension directe de cette contribution à la reconnaissance et la localisation de classes d'objets. Dans un premier temps, nous avons développé une méthode inspiré de l'appariement de graphe (i.e. graph matching) afin de traiter le problème de la reconnaissance rapide d'instances d'objets spécifiques dans des conditions bruitées. Cette méthode permet de rajouter facilement un nombre quelconque d’autres types de caractéristiques locales (e.g. contours, textures…) moins affectées par le bruit tout en contournant le problème de la normalisation et sans pénaliser la vitesse de détection. Nos expériences sur plusieurs bases de test ont montré la pertinence de notre approche. Notre approche est globalement légèrement moins robuste à l'occultation que les approches existantes, mais elle produit des performances supérieures aux approches standard en conditions bruitées. Dans un second temps, nous avons développé une approche pour la détection de classes d'objets dans le même esprit que celui du sac de mots visuels. Pour cela, nous utilisons nos cascades de micro-classifieurs pour reconnaître des mots visuels plus distinctifs que les mots basés simplement sur des points d'intérêts. L'apprentissage se divise en deux parties: dans un premier temps, nous générons des cascades de micro-classifieurs servant à reconnaître des parties locales des images modèles ; puis dans un second temps, nous utilisons un classifieur afin de modéliser la frontière de décision entre les images de classe et celles de non-classe. Nous montrons que l'association de mots classiques (à partir de points d'intérêts) et de nos mots plus distincts produit une amélioration significative des performances pour un temps de calcul assez faible. / In this thesis, we first present a contribution to overcome this problem of robustness for the recognition of object instances, then we straightly extend this contribution to the detection and localization of classes of objects. In a first step, we have developed a method inspired by graph matching to address the problem of fast recognition of instances of specific objects in noisy conditions. This method allows to easily combine any types of local features (eg contours, textures ...) less affected by noise than keypoints, while bypassing the normalization problem and without penalizing too much the detection speed. Unlike other methods based on a global rigid transformation, our approach is robust to complex deformations such as those due to perspective or those non-rigid inherent to the model itself (e.g. a face, a flexible magazine). Our experiments on several datasets have showed the relevance of our approach. It is overall slightly less robust to occlusion than existing approaches, but it produces better performances in noisy conditions. In a second step, we have developed an approach for detecting classes of objects in the same spirit as the bag-of-visual-words model. For this we use our cascaded micro-classifiers to recognize visual words more distinctive than the classical words simply based on visual dictionaries. Training is divided into two parts: First, we generate cascades of micro-classifiers for recognizing local parts of the model pictures and then in a second step, we use a classifier to model the decision boundary between images of class and those of non-class. We show that the association of classical visual words (from keypoints patches) and our disctinctive words results in a significant improvement. The computation time is generally quite low, given the structure of the cascades that minimizes the detection time and the form of the classifier is extremely fast to evaluate.