Video Summary Key Technology

With the rapid development of communication technologies, broadband network technologies and digital TV, video data is showing explosive growth.However, unstructured data in the form of video information to video browsing efficiency is very low.Nowdays,it’s need to be urgently addressed about how to organize and manage these vast amounts of video information in order to achieve fast, effective search。 The birth of video summarization is in this case, its purpose is to improve the efficiency of browsing in order to reduce the amount of video data processing, it is very important for a variety of video-based application. Video abstraction has become a hot research topic.The paper first introduces the video abstraction of research background and research status, characteristics and structure of the video data。 Based on this, it elaborates video abstraction technology, provides an overview of existing video segmentation and key frame extraction algorithm to analyze the advantages and disadvantages.In aspect of video segmentation, This paper presents a segmentation algorithm based on the lens of the adaptive dual-threshold. By contrast to the HSV and RGB two color model advantages and disadvantages, the HSV model color histogram is established, and quantified to the characteristics of the video frame, to calculate the inter-frame similarity.Through the comparison of the inter-frame similarity and dual-threshold, It able to detect the mutation and gradient at the same time. Adaptive threshold combining with dynamic sliding window can be obtained by computer calculation, Can fully adapt to the local frame changes, and achieved good results during the video segmentation.Based on the video segmentation,using cluster technology, this paper presents a lens cohesion-based hierarchical clustering algorithm to extract the key frames. First image block and give a different coefficient values, and then calculate the inter-frame similarity.This idea of sub-blocks taking the video content in every frame into account, it can more accurately the similarity between each frame. Calculate the adaptive threshold value in accordance with the similarity, the shots of video segmentation as the initial class.Next,comparing threshold value and the distance between classes,it takes agglomerative hierarchical clustering baseing on shots.This way reduces the redundancy between the shots.Then, the key frames extracted from each class.Finally,this paper has designed and developed a video abstraction system of a human-computer interaction,achieving two basic functions of the video summary system---video segmentation and key frame extraction. Also,the effectiveness of the proposed algorithm is verified by experimental analysis.