Abstract:
This Dissertation report discusses one of the computer vision problem, which is implemented using various video and image processing techniques from last two decades that is “TEMPORAL VIDEO SCENE SEGMENTATION”. From last decade a lot of work has been done on automatically video scene segmentation. In field of computer vision this problem has received most popularity because it is the first and most important part of the other problems like video summarization, video indexing and browsing etc. This dissertation includes a new approaches to solve this problem which consists of different feature detector and descriptors, clustering algorithms, window based scene boundary defining etc. As we know scenes are the grouping of semantically similar shots which are temporarily close. So our first task is divide the video into shots, which includes cluster of similar frames. This task is done using HSV color histogram. Now we extract some visual, motion and SIFT features of the shots and calculated inter shot pairs similarity, which is interpreted with shot similarity graph (SSG). At the final stage using sliding window method we have grouped similar shots which are higher similarity then some decided threshold. This grouping has considered inverse time proximity. This dissertation has also include some of implemented papers with their different approaches and achieved results are also discussed.
