Based Key Frame Near-duplicate Video Retrieval

With the popularity of smart phones, digital cameras, other image acquisition devices and the development of the multimedia technology, mass near duplicate video appeared on the Internet media and social networks. Near duplicate video retrieval become one of the focus areas of multimedia technology. In order to improve the near duplicate video retrieva’s timeliness and accuracy, this dissertation do the following research works:1. A key frame extraction algorithm is presented, which is based on statistical average gray value of the image. The method first obtains the average gray value of the image, and get the extreme points by the gray value difference. Then matching the adjacent image to filter the high degree of similarity image. The basis of taking the average gray value of the image as the key fame is that it has strong robustness to noise. The experiment results show that the method can extract the key frame better.2. An improved multiple feature hash near duplicate video retrieval algorithm is proposed. The method improves the adjacency matrix elements. The elements of the adjacency matrix is longer use 0 and 1 two values, but use Gauss kernel function. The method improves the retrieval accuracy of the multiple feature hashing, and the experimental results show that it can improve the precision of near duplicate video retrieval.3. A survelliance video retrieval and positioning algorithm is proposed. The method takes advantage of key frames’s hash code to search the key frame in long video which is similar to the short video’s first, and end frame, extract the key frames between two frames using hash code to match the short video, find the similar video clips, so that the short video locating in the long video is realized. This method takes advantage of the inherent time video information, improve the precision of retrieval.