Dissertation > Excellent graduate degree dissertation topics show

Research on the Technology of Scene Summarization in Video Retrieval

Author: LiGuangCui
Tutor: ZhangJianMing
School: Jiangsu University
Course: Applied Computer Technology
Keywords: Scene Summary Swarm intelligence clustering algorithm k-means algorithm Scene Segmentation Lens similarity Visual attention model
CLC: TP391.41
Type: Master's thesis
Year: 2010
Downloads: 64
Quote: 0
Read: Download Dissertation

Abstract

With the rapid development of multimedia technology, multimedia data is increased exponentially, and how it can be effectively stored, segmentation and retrieval has become pressing issues that need to be addressed. Video summary technology is a way to solve the problem. At the same time, the video summary also assisted the establishment of video retrieval and indexing system. Therefore of great significance for the study of video summary. In this paper, a comprehensive analysis has been based on the basis of the content of the video summary of swarm intelligence combined with the k-means-based key frame extraction algorithm, based on the characteristics of the camera movement and time information of the scene segmentation method based on visual attention model The scene summary generation method, designed and implemented a prototype system. The details are as follows: (1) a combination of swarm intelligence and k-means-based key frame extraction algorithm. By introducing a clustering algorithm based on swarm intelligence, and k-means clustering algorithm combination applied to the key frame extraction. The first clustering algorithm based on swarm intelligence organizations to form an initial cluster, perform the k-means algorithm, the results of the initial cluster clustering of optimizing initial results and to speed up the convergence of the algorithm, the final extract on behalf of the entire lens of the key frames. Traditional clustering method to extract key frames to effectively solve the initial value sensitive, prone to premature and requires the use of a priori knowledge to determine the cluster number of categories. (2) propose a scene segmentation method based on the characteristics of the camera movement and time information. Traditional scene segmentation method using only visual features the scene segmentation caused by the scene split inaccurate, considering the visual characteristics of the lens, lens similarity measure, the time information included in the global motion characteristics and lens, and the lens chain clustering algorithm similar to the lens together to the same scene, scene segmentation to achieve better effect. (3) Abstract generation method based on visual attention model scene. Visual attention model to the the summary scene generation process, through the extraction of lens movement significantly with the degree and static significant degree, the movement significantly with the degree and static significant degree of visual note that the degree of combination to get the lens; at the same time, considering the duration of the lens on the lens important degree, degree of sustained the lens visual attention and time weighted summation, the importance of the lens. Selected according to the size of the lens degree of importance scenes important lens, arranged chronologically through the lens keyframe output, generates the summary of the scene of the scene to main content. (4) the use of object-oriented design and development scene summary generated prototype system. The system includes shot boundary detection, key frame extraction, scene segmentation, scene summary generation module. And compared through experiments to verify the validity of the above-described method.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile