News video summarization due to changes of image features

In collaboration with developments on multi-media technologies, video usage increases in daily life. In case of necessity, searching can be done on data which is recorded on video. Because of being hard of searching and getting results within huge-sized data, researching about video summarization gets importance nowadays. Video summarization is an action which determines scene changing and keeps them during image flow. In this study, a system which summarizes videos about news, was designed. Videos about news are video types which on searching can be done mostly and have the most scene changing. Summarization of a video about news has to inform users about general concept of the video with least interference. It is important because of getting results quickly when user wants to search data on the video. In this study, color, edge and texture features are examined while video flow is being summarized. Firstly, color feature which is basic feature of scene and examined quickly in terms of performance, is examined with 64-bin histogram method. Edge feature is examined when color change is higher than a certain threshold value by using edge changing ratio method. Texture feature is examined when edge change is higher than a certain threshold value by using Gabor filter with 30o angle differences between 0o and 180o angles. Performance of this method is bad in terms of time interval. Therefore, in order to increase general performance of the system, last examination is done on texture feature. With this video summarization study, 25 news videos which have different contents were summarized. Summarized videos are shown to 40 users and wanted users to mark scenes that they think about insufficient or extra. Comments of users are examined by using Mean Average Precision method and in order to results of the method, it is seen that application has a success ratio of %98. Key Words : Video summarization, video archiving, 64-bin histogram, edge change ratio, Gabor filter bank.