The recognition of object categories has become a major topic in computer vision. Accurate automatic visual recognition is central to applications such as image search, robotics, and vehicle safety systems. This book proposes new discriminative learning approaches to three recognition tasks. 'Image classification' aims to determine the presence of an object of a particular category in an image or video frame. 'Object detection' further aims to localize these objects. And 'semantic segmentation' aims to partition an image into coherent regions of particular categories. First, we show that a codebook of learned fragments of object contours is sufficient for accurate recognition. Second, we propose new texture features that simultaneously exploit local appearance, approximate shape and appearance context. The efficacy of the new contour and texture features is tested on a variety of challenging image datasets. Finally, we show how a combination of these two largely orthogonal features improves recognition above that achieved by either feature alone.

Данное издание не является оригинальным. Книга печатается по технологии принт-он-деманд после получения заказа.