Evaluation of the clustering of video frames using Rank and Histogram methods with Euclidean and City Block distance measurement for different levels of threshold

In this work, we present the results of evaluating the clustering of video scenes. To evaluate the clustering we have developed a technique to detect changes in the information that is present in the video frames. For clustering, we measure the distance between features of two consecutive frames to decide if a frame belongs to a cluster. The threshold was settled with different values during the experimentation. We used the Rank and the Histogram as frame features, and Euclidean and City Block for the distance measurement. Tested videos are those from the MPEG-7 Content Set with different lengths, frame sizes and frame rates that serves as reference for the measurement. For the selected threshold, we present the best combinations to get the best results, showing that histogram method present better outcomes.