Recognizing People in Images and Videos

Abstract

Cameras and mobile phones have become integral part of our everyday lives as they become portable, newlinepowerful and cheaper. We capture and share hundreds of pictures and videos with our friends, family newlineand social connections. Similarly, large volume of such visual content is generated in surveillance, newlineentertainment, and biometrics applications. Without any doubt, people are the most important objects newlinethat dominate in these visual content. For instance, photos taken in a family event or movie videos focus newlinearound humans. It is utmost important to automatically detect, identify and analyze people appearing in newlineimages to obtain a better understanding of these content and make decisions around them. newlineIn this thesis, we consider the problem of person detection and recognition in images. This is a newlinewell explored topic in vision community with a vast literature focused on these problems. The current newlinestate-of-the-art recognition systems are able to identify people with high degree of accuracy in scenarios newlinewhere images have high resolution, contain visible and near frontal faces, and recognition systems have newlineaccess to sufficiently large training gallery. However, these systems need significant improvement in newlinechallenging real-world applications such as surveillance or entertainment videos where one needs to newlinehandle several practical issues such as non-visibility of faces, limitation of training samples, domain newlinemismatch, etc in addition to other instance variations such as pose, illumination, and resolution. While newlinethere are plenty of challenges pertaining to person recognition, we are interested in some of the open newlinechallenges that are relevant from the deployment perspective in diverse recognition scenarios. newlineWe first consider people detection that is a pre-requisite for a recognition system. We detect people newlinein images by detecting their faces through an exemplar based detector. Exemplar approach detects faces newlinethrough hough voting using an exemplar training database indexed with bag-of-words method. We propose two key ideas referred