Visual detection and recognition using local features

Computer Vision;Image Representation;Object Detection;Object Recognition;Parallel Programming;GPU Programming;graphics processing unit (GPU)

Dikmen, Mert

关键词: Computer Vision;    Image Representation;    Object Detection;    Object Recognition;    Parallel Programming;    GPU Programming;    graphics processing unit (GPU);   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/32069/Dikmen_Mert.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Detection and recognition of objects in images is one of the most impor-tant problems in computer vision. In this thesis we adhere to a traditionalbottom–up detection and recognition framework, where the objects are firstlocalized with a sliding window detector before being identified. We makemultiple contributions along this path. All of the contributions pertain tothe central theme of local image features.We demonstrate improved object detection performance with our proposedfeature extraction process, which generalizes the traditional feature extrac-tion methodology of pooling atomic appearance information (e.g., image gra-dients) around pixels in localized histograms. In addition, we propose amethod to fuse two types of information sources in a locally discriminativemanner by leveraging local class-dependent correlations.For the recognition task, we adopt a state–of–the–art metric learningmethod and modify it to handle unknown identities.Lastly, the computational improvements achieved through leveraging par-allelism are brought together by the Vision Video Library (ViVid), which werelease as open source to the research community.