A DCT coefficient sign-based background model for moving objects detection from motion JPEG coded movies

This paper proposes a robust moving objects detection method for motion JPEG coded movies. The proposed method utilizes the similarity based on of the positive and negative sign of discrete cosine transformed (DCT) coefficients. The proposed method describes a time-varying background by an adaptive model consisting of multiple sets of DCT signs with respect to each 8 × 8 DCT block to adapt to non-stationary scenes. In addition, the multi block-based processing reduces false detection. Moreover, since DCT sign is encoded separately from its corresponding magnitude in a motion JPEG codestream, the sign can be obtained without decoding a codestream. Experimental results comparing with a conventional method using Gaussian mixture model show the effectiveness of the proposed method.