Object recognition in images is one of the oldest problems in Computer Vision. In this thesis, we focus on the problem of category-level object recognition, based on the use of shape features as a more generic representation of object classes than appearance features, while the latest are used as second-level features. In this research work we propose an invariant shape feature extraction, description and matching method for binary images, named LISF. The proposed method extracts local features from the contour to describe shape and these features are later matched globally. Combining local features with global matching allows us to obtain a trade-off between discriminative power and robustness to noise and occlusion in the contour. The proposed extraction, description and matching methods are invariant to rotation, translation and scale, and present certain robustness to partial occlusion. The conducted experiments showed that, for larger occlusion levels, the better was the performance of LISF with respect to other popular shape description methods, with about 20% higher bull's eye score and 25% higher accuracy in classification in images with a 60% occlusion. Also, we present a massively parallel implementation in GPU of LISF, which achieves a speed-up of up to 34x. In order to deal with the intrinsic problems derived from using edges extracted from real images, in this thesis we propose a shape descriptor, named OCTAR, that is particularly suitable for partial shape matching of open/closed contours extracted from edgemap images. Based on this descriptor, we also propose a partial shape matching method robust to partial occlusion, noise, rotation and translation. Our approach allows to combine shape and appearance through the evaluation of object detection hypothesis based on its appearance, providing more distinctiveness. The conducted experiments showed competitive results compared to the state of the art, in both binary and gray scale images.