Video sequence matching

Abstract

With the rapid growth of IT technology, the use of multimedia content such as images and videos is growing explosively. Multimedia data is available everywhere in our daily lives. The popularity and rise of social networking has played a huge part for this rapid growth of multimedia content. Over one hundred thousand of videos are uploaded to the Internet every day and those videos are played over fifteen billion times per day. According to the latest Cisco visual networking index forecast of 2016, three-fourths of the world’s mobile data traffic will be video by 2020. Such growth in multimedia content requires the development of techniques to manage and access these large volumes of visual data. The fundamental problems in the retrieval of visual data such as videos are the design of good data representations and a definition of a quantitative metric that efficiently measures the similarities between each pair of visual data. A video is none other than a sequence of images. There are numerous application areas where we need to compare two long sequences and to measure their similarity. Bioinformatics and error-correction are two widely known applications that make use of such algorithms. Sequences may differ in symbol substitutions, insertion of extra symbols, deletion of symbols or possibly transposition of adjacent symbols. The goal of this dissertation is video sequence matching system to establish a form of similarity measure based on an established ground truth assuming there was no intensity or colour changes to the query video independent on whether the sources of the two videos being matched are extracted from the same source or different sources. The implemented system is based on intensity histograms of video frames. The system best identified a form of similarity using the Chebyshev distance and the ideal upper thresholding values. The system produced an accuracy ranging from 28% to 71%. The system fails when there are any colour or intensity changes, but overall the aim to find a form of similarity whether being an exact match or a partial match has been reached.