An improved background modeling algorithm based on the codebook model

As an effective background subtraction algorithm, the basic codebook model algorithm has many disadvantages. Such as too many parameters, background was mismatched as foreground, the slow speed of the background model updating, affected by noises. An improved codebook model algorithm is proposed in this article. The mean and variance cubic (MVC) of the pixel value in RGB color space was applied to judge whether the input matches one of the codewords or not; the mean and variance of codeword is updated by dynamic learning rate which is correlated to the frequency of the codeword has occurred; Layered modeling technique was proposed to update codebook adaptively; meanwhile the frequency of the codeword has occurred was also used as one of the criteria to judge whether absorbing a codeword into the codebook model. Experiment results show that the algorithm proposed can segment moving objects under complicated scenes which could include dynamic background and illumination change, can absorb variations into the background model quickly, can be used to real-time foreground-background segmentation.