Research On Video Retrieval Key Techniques Based On HEVC Compressed Domain

At present,people spend a lot of effort on the key technologies of video retrieval technology in the pixel domain.Among them,there are many key frame extraction techniques,feature vector extraction techniques and feature matching techniques.Although some scholars have researched compressed domain video retrieval based on the early video coding standards,they have applied keyframe extraction techniques,feature vector extraction techniques,and feature matching techniques in the pixel domain,failing to respond to the corresponding video formats.Processing.This paper aims at the HEVC compressed domain video stream,and analyzes the latest HEVC video coding standards,including coding units and intra prediction based on the HEVC compressed domain code stream,to perform key technologies for video retrieval based on HEVC compressed domain.the study.The research work of the thesis is mainly divided into three parts:First,key frame extraction.By analyzing the I-frame stream in the compressed video stream in HEVC domain,using the energy concentration characteristics of DCT coefficients and storing and fast and efficient for subsequent video retrieval,the DC coefficients is used to extract the keyframe I frame.Experiments show that the proposed algorithm is more effective in extracting keyframe I-frame images.Second,feature vector extraction.On the premise of the extraction of I-frame of key frame in the third chapter,using the intra-prediction mode information and prediction unit size information of HEVC compressed domain I-frame image,this paper proposes a feature extraction algorithm based on intra-prediction mode and prediction unit size.The algorithm combines the relationship between the I-frame and the image texture,and uses the prediction mode histogram and the prediction unit size histogram of the luma component,the chroma component in the I-frame as feature vectors.And the histogram is normalized to reduce the impact of the error of one component feature as much as possible,and greatly reduce the influence factor caused by a certain feature component.Experiments show that the proposed feature extraction algorithm can extract video image features better.Third,feature matching.In the third chapter,the I-frames have been extracted by using the concentration of DCT energy.In the fourth chapter,the feature vectors of I-frames are extracted by the feature vector extraction technology based on intra-frame prediction mode and prediction unit size.In order to complete the video retrieval function,it is necessary to use a suitable matching algorithm.For this reason,a feature matching algorithm based on intra prediction mode and prediction unit size is proposed.The algorithm makes full use of the characteristics of the extracted eigenvectors: the increase of intra-prediction mode and the change of prediction unit size to judge the similarity between images,so as to realize the function of matching retrieval.Experiments show that the proposed algorithm can complete the matching function better.