The theoretical and the experimental investigations of the laser polarization imaging system of speckle statistical characteristics and speckle removing method are conducted. A probability distribution model for speckle intensity of laser active polarization imaging system by using Müller matrix is proposed. According to modern charge coupled device (CCD) imaging characteristics, through comparison of speckle and pixel size speckles are divided into two kinds, namely small speckle and big speckle. The normalized variance of speckle is linear to pixel, and the mean value of speckle is irrelevant to pixel. A unified speckle noise probability model, namely the improved Gamma distribution model is proposed. Based on this model, the improved Bayesian nonlocal algorithm is presented. The despeckling results show that equivalent number of looks and edge preserve index outperform those of traditional methods.