Video Hashing Based On Motion Information

For video copyright protection, common technology is digital watermarking and content-based copy detection technology. The former is replaced gradually by the latter because inconvenience. Content-based video copy detection is a technology of video content enrichment that is mainly used to detect whether there is the illegal operation of video copy, etc. The main idea is extracted some features that can represent the characteristics of the video content from a video streaming, and measure the distance of the features and other features extracted from some copies in the library. Finally a conclusion will be given. Feature representation closely are linked to the accurate, strategy and efficiency of detection algorithm. Perceptual hash is to compress and map the perceptual characteristics of a video to a short bit string or a string. Similar objects generated same hash values. Different objects produce different hash values. Perceptual hash can be used to represent video features because the uniqueness of the hash values ensures the similarity between different content of video is extremely low, robustness to ensure the same high similarity between the video content. This paper mainly studies how to extract excellent features from a video, and map it to the hash values. This paper introduces the content-based video copy detection technology and perception of the hash key technology. Two video copy detection algorithms based on hash are given through a large number of experimental verification. Specific research results are illustrated as follows.1. A video perceptual hash technology method based on motion information is proposed.The steady motion information are extracted from two adjacent frames. It represents a part of the video content changes. A frame can be divided into 2n×2n pieces. The size and location of the motion information will represent the features of a frame. If a motion information covered a piece, the piece value will be marked. Finally, the piece values form binary hash code by connecting the Hilbert curve. The ratio of the number of same squares and own squares can measure the similarity of two key frames. The similarity of two videos is measured by the ratio of the sum of key frame similarity and the numbers of total key frame.2. A video perceptual hash method based on motion information and SURF is proposed.Because motion information does not recognize the local information of a video frame image, SURF local characteristics are introduced. After extracting motion information the key frame image have removed many background information. The SURF feature points are extracted in the image. A Frame can be divided into 2"x2" pieces. If the coordinates of SURF feature points is the same as the piece coordinates, the piece value will be marked. The piece values form binary hash code by connecting a m-layers of closed circular curve. The calculation of each key frames similarity and the entire video similarity are given. Characteristics of SURF rotation invariance ensures that no matter how to flip the image feature points detection in the same place. The proposed algorithm can solve the video flip attack operation.The hash code method based on the m-layers of closed loop curve starts is proposed. Start from the center of the 2n×2n small squares, according to a fixed order, such as the upper right corner of the square, the block values are extracted to form a circle number, m lap m loop code. The method can solve the problem of feature points moving along the frame center because the rotation the image,such as flip horizontal. When matching the hash codes, a m-levels matching algorithm has been proposed.With a large number of test cases, this article tests the two methods, the determination of the original video and copies of video similarity. Experiments shows that the above two methods showed a better robustness and distinction against most of the video attack in this test.