Adaptive transform domain video scene analysis

An adaptive mechanism for video sequence analysis in the transform domain is proposed. Methods for measuring motion complexity and activity levels in the scene are presented, based on which a video scene can be consistently categorised into identifiable classes. Further, the video quality, as measured by the mean square error or the signal to noise ratio is related to certain parameters used in video analysis. Adaptability in the analysis is realised by tailoring the parameters of the video analysis algorithm to the specific characteristics of the video scene, as embodied in the video scene class and the video quality. Preliminary experiments show the feasibility of the proposal, and the improved performance that can be achieved by adaptively tuning the analysis parameters to the characteristics of the particular scene under consideration.