| Summary: | 碩士 === 國立臺灣科技大學 === 資訊工程系 === 90 === Recently, Ardebilian et al. presented a scene change detection technique us-ing the focus of expansion points which are the intersections of the detected lines in an image frame. They used an adaptive threshold method to de-termine the threshold for detecting scene change. Given a video sequence, this paper first presents a randomized shape—analysis algorithm for detect-ing shapes, such as lines, circles, and ellipses, within the predicted search areas which are found in the reference image frame. For each current image frame, the shapes within the predicted search areas can be detected in a real—time manner. After detecting all the shapes in each image frame, all
the intersections between these shapes are obtained and these intersections are representative of the image frame. Further, using these intersections of the reference image frame and the current image frame, the partial Hausdorff distance criterion is used to measure the dissimilarity between the reference
image frame and the current image frame in order to check whether there is a scene change or not. Under a real video sequence, experimental results re-veal that our proposed scene change detection algorithm is quite competitive with the previous algorithm proposed by Ardebilian et al.
|
|---|
