Font Size: a A A

Research On Video Image Content Matching And Retrieval

Posted on:2018-12-16Degree:MasterType:Thesis
Country:ChinaCandidate:Y PiFull Text:PDF
GTID:2348330542469900Subject:Electronic Science and Technology
Abstract/Summary:PDF Full Text Request
With the development of multimedia technology,multimedia carrier form has changed from the original text to the image and then to the video.In the face of massive video information,how to effectively analyze and retrieval it has become a huge challenge.At present,content-based video retrieval technology development has made great achievement compared to traditional text-based video search technology performance has been greatly improved.Content-based video retrieval technology generally uses the underlying visual feature of the key frame or the lens frame to express the video content,and the result of the video retrieval is determined by calculating the similarity between the query image feature and the key frame feature.In the actual video retrieval,the user's final demand is to be able to directly retrieve the object in the video,so object-based video retrieval is the main trend of future development.This paper studies the object-based video content retrieval technology,which mainly includes the research of object extraction in video and the object-based image retrieval research.The purpose of extracting objects in a video is to create object library information for the video and to provide retrieval data for subsequent object retrieval.The method of object detection in video can be divided into two categories:dynamic target detection and static image target detection.The dynamic target detection method will lose the detection information for the target in the image which still for a long time,and it will be judged as background.For static image target detection method,due to motion blur,the object area is too small,and it will miss some of the target information.In order to solve the above problems,this paper combines two methods to complete the video object extraction task,and through the experimental analysis,the fusion method can improve the detection rate of the object in the video.After the establishment of the video object library,the next step in the study is the problem of object-based image retrieval.Image retrieval task is through similarity analysis between the given query picture and image library image,to find a similar image.Image retrieval includes image feature extraction and similarity metrics.the effect of consequence retrieval is largely determined by the performance of image feature extraction.Image recognition based on depth learning has made great progress in recent years,but the research work in the field of image retrieval is less.This paper based on the deep convolution neural network,assistant network,and the merged characteristics of the multiple convolution layers as the features of the image to improve the disadvantage of traditional manual design features in image retrieval tasks.Through the experimental analysis,the method of the extracted features proposed on this paper perform better compared to the feature of original depth convolution neural network and the traditional features on image retrieval tasks.Finally,based on the above research,an object-based video retrieval system is designed and implemented to provide users with basic video retrieval services.
Keywords/Search Tags:Video Retrieval, Deep Learning, Object Detection, Feature Extraction, Image Retrieval
PDF Full Text Request
Related items