Adaptive thresholding for scene change detection

Increasing the demands of digital video by developing the Internet market and the multimedia technologies, video indexing technique such as scene change detection is required to manage the data efficiently. In conventional methods, scene change is detected by comparing the value of current detected measure with the fixed threshold induced from preceding experiments. However, this can not guarantee the best performances on all various video sequences, due to their own specific characteristics. To solve this problem, a novel adaptive threshold decision method is proposed. First, histogram of scene change detection measure induced from preceding experiments is derived. Then this is modeled with log-normal distributed pdf and the model parameters are estimated. Consequently, experimental results obtained from this pdf model and estimated parameters demonstrate better performance of the proposed method comparing with conventional methods.