A video tensor self-descriptor based on variable size block matching

Abstract

This paper presents a different and simple approach for video description using only block matching vectors, considering that most works on the field are based on the gradient of image intensities. We first divide the image into blocks of different sizes. The block matching method returns a displacement vector for each block, which we use as motion information to obtain orientation tensors and to generate the final self-descriptor, since it depends only on the input video. The resulting descriptor is evaluated by a classification of KTH, UCF11 and Hollywood2 video datasets with a non-linear SVM classifier. Our results indicate that the method runs fast and has fairly competitive results compared to similar approaches. It is suitable when the time response is a major application issue. It also generates compact descriptors which are desirable to describe large datasets.