Scene Change Detection using Different Color Pallets and Performance Comparison

Abstract

In the world of massive uploaded videos, to be able to cover the content of a video at a glance becomes a necessity since there is no enough time to watch the whole video for an individual. Looking at frames of different scenes in a video gives a brief idea of the content, when each different scene images are listed to be checked by the user. In this study, an approach using various color palettes is proposed to be able to detect the different scenes of a video. In the proposed method, color histogram values of sequential frames firstly are calculated. If the difference in the histogram values of the pair frames in sequence is over a threshold value (percentage of change), scene change is detected. In the experimental studies, 3-Bit RGB (Red Green Blue), 6-Bit RGB, 8Bit RGB, 9-Bit RGB, 1-Bit Binary, 4-Bit Gray, and 8-Bit Gray palettes are implemented over a list of video files and compared. In the comparisons of palettes, accuracy, precision, recall, and F1Score performance metrics are used. In the performance accuracy controls, 6-Bit RGB color pallet with a threshold level value of 35% has been experimented as the best of all.