Non ti piace? Non importa! Puoi restituire gli articoli fino a 30 giorni
Non puoi sbagliarti con un buono regalo. Con il buono regalo, il destinatario può scegliere qualsiasi prodotto della nostra offerta.
Fino a 30 giorni per il reso
The recognition of object categories has become a major topicin computer vision. Accurate automatic visual recognition iscentral to applications such as image search, robotics, and vehiclesafety systems. This book proposes new discriminative learningapproaches to three recognition tasks. \'Image classification\'aims to determine the presence of an object of a particularcategory in an image or video frame. \'Object detection\' furtheraims to localize these objects. And \'semantic segmentation\' aimsto partition an image into coherent regions of particularcategories. First, we show that a codebook of learned fragments ofobject contours is sufficient for accurate recognition. Second, wepropose new texture features that simultaneously exploit localappearance, approximate shape and appearance context. The efficacyof the new contour and texture features is tested on a variety ofchallenging image datasets. Finally, we show how a combination ofthese two largely orthogonal features improves recognition abovethat achieved by either feature alone.