Research On Feature Extraction Based On ISA Of Video Object

In the research and applications of image and video,computer vision and other technologies have been continuously integrated into the image and video object recognition and analysis methods.However,due to the rich of content,the variety of objects,the complex background,and the large amount of noise in images and videos,there are still many challenging issues in the identification,detection and analysis of image and video objects.At the same time,how to make efficient feature extraction and convenient recognition of objects in increasingly rich image and video information has become an urgent problem to be solved.In this thesis,we combine the related research knowledge in recent years to study the problem of image feature extraction and classification in images and videos.Experiments show that the proposed algorithm in this thesis improves the accuracy and robustness of objects classification and recognition in images and video.The main contents of this thesis are as follows:(1)Elaborating the relevant object feature extraction method and its principle and characteristics.The SIFT method of image local feature extraction,the sparse coding inspired by the visual perception principle in the human visual system,and the independent subspace analysis(ISA)with the ability to simulate the cell perception in the V1 region of the human visual system are introduced in detail.(2)Through the research and analysis of various feature extraction algorithms,this thesis selects independent subspace analysis(ISA)method.The process of ISA model applied to the image is described in detail and the ISA model is used in image scene classification and recognition experiments,and compared with other feature extraction methods.The experimental results verify the effectiveness and accuracy of the ISA method in objects classification and recognition in images.(3)Combine the original ISA model with the deep neural network theory to form an ISA deep network model,and elaborate on its processing of image and video data.The ISA deep network model was experimented on the Hollywood Human Behavior Data Set to identify and classify the human behavior in the video.It was proved that the behavior recognition method based on ISA deep network model can learnsignificant features with good expressive ability,and obtain good behavior recognition accuracy.