Investigation Of Refinement Means For Microstructure Of Hypereutectic Al-Fe Casting Alloys

Today, as times progress and development of technology,The video shows all aspects ofthe society and life.How to retrieve and classify the video information has become an urgenttopic.In order to get useful information from video media library effectively,effective organizat-ion and indexing of video information is necessary. Therefore, the research of content-basedvideo retrieval and classification has gradually become a hot research topic in the field ofmultimedia technology.This paper summarizes the characteristics of video data, video structuring and key techni-ques in content-based video retrieval and classification and places emphasis on study of shot de-tection and key frame extraction. Based on the former research of video segmentation, this pap-er proposes a new algorithm based on adaptive double threshold. Comparing with the formeralgorithm, new algorithm reduces the mutation and gradient threshold, improves the recall ratioand reduces the error by using partitioning strategy which has different weight and eliminatingpart of a larger difference between frames. Based on the former research of key frame extraction,this paper also puts forward an improved video key frame extraction algorithm which based onmutual information.The algorithm optimizes the number of key frames, makes the number ofkey frames can adjust automatically according to the video content, increases the number of keyframes of adaptability, finally makes key fra-mes which has been extracted can descript thevideo content better.The experime-ntal results show that key frames extracted by the algorithmcan effectively sum up the content of the video and that the algorithm improves the extractionefficiency and reduces the key frame redundancy. In this paper, the feature extracted isprocessed by dimension reduction, the paper employs genetic algorithm to optimize the SVMclassifier parameters.For finding out the more accurate classification feature subset, the paperdecreases feature dimension and optimizes the color features on the premise of identificationaccuracy. At the same time, using the optimization of the parameters of the classifier selection to improve the classification accuracy and classificat-ion speed.The system of video retrievaland classification was designed in the form of modular-ization. the every module has beendepicted in detail.The experimental results show that by using the new Shot boundary detection algorithm onthe test set, the average recall and precision are reached87.86%and93.91%respectively. Thenew shot boundary detection algorithm has a better effect. The new key frame extractionalgorithm extract the21video frame for key frames from animation which has12470framesand115shots and the former algorithm extract the13video frame for key frames fromanimation which has605frames and11shots.The data shows that key frames extracted by thealgorithm can effectively sum up the content of the video and that the algorithm improves theextraction efficiency and reduces the key frame redundancy.Finally this paper summarized the research work and put forward the direction of next step.