A Compressive Sampling Approach to Scene Change Detection

As an initial step of intelligent video processing, the efficiency of scene change detection algorithm mostly impacts the whole processing progress. Increasing number of researchers has absorbed in improving the algorithm's accuracy or reducing the computational complexity through various approaches. This paper proposes a new approach to scene change detection which explores the high compression efficiency of compressive sampling theory, and only uses 20 compressive samplers each frame to achieve a high scene change detection accuracy. The main contributions of this paper are as follows: we first introduce the compressive sampling theory into video scene change detection technique, then found the best combination of sampler number and the threshold so as to achieve the highest accuracy, and importantly the computational complexity of the proposed algorithm keeps at a low level. The experimental results demonstrate our algorithm's efficiency: 0.93 of precision rate, 1.0 of recall rate and 0.96 of F1 score on average. Meanwhile, comparing with the algorithm with the similar computation complexity, the proposed algorithm is able to improve the detection accuracy by approximate 20%, comparing with the algorithm with the similar detection accuracy, the proposed algorithm can reduce approximate about 40% of computational time.