Performance Evaluation of Algorithm for Detection of Fades in Video Sequences in Presence of motion and Illumination

Abstract

Abstract. The detection of a fade in and a fade out in the video segmentation is very difficult as compared to the detection of abrupt transition. Mostly, the disturbances are caused due to the object/camera motion and the illumination arising from flashlight, fire, flicker and explosion. This mainly results into false positives. An algorithm has been proposed for the fade in and fades out detection. In the proposed algorithm, frames are decomposed to 5 levels using db-6 discrete wavelet transform. The mean of approximate wavelet coefficients are calculated for each frame and normalized. The variation in normalized mean of wavelet coefficient against frame index gives the presence of fade in and fades out. In our algorithm, no preprocessing or post processing is required and so the computational complexity is reduced to a great extent. The processing time is very small as compared to the previous techniques. The proposed algorithm detects and identifies almost all the cases of fade in and fade out with a negligible miss and the false detection causing due to illumination and motion. Performance comparison of the proposed algorithm with the other existing techniques clearly demonstrated its effectiveness in terms of performance metrics such as Recall, Precision, F1 measure, RSI and detection rate. Another advantage of our algorithm over the previous ones is that the detection rate is almost 100%.