An efficiently adaptive probability estimation method in CABAC of HEVC using stochastic learning weak estimator

In the coding process of Context-based Adaptive Binary Arithmetic Coding (CABAC) in High Efficiency Video Coding (HEVC), the probability estimation with table lookup is adopted to improve the speed of the estimation at the cost of the compression efficiency. In order to improve the compression efficiency while maintaining the speed, an efficiently adaptive probability estimation method based on the Stochastic Learning Weak Estimator (SLWE) is proposed. Due to the strong adaptability of SLWE to non-stationary data, a basic framework of probability estimation with SLWE is firstly established by multiplication principle on the learning factor instead of look-up table operation. To further enhance the compression efficiency, a new learning factor with the expression of 1-1/2w is designed benefiting from its adaptability to data characteristic and the form of power of 2. Thus, the next estimation probability can be computed by simple operations in place of multiplication, including binary shift of the current symbol’s probability and addition or subtraction for the values resulting from the shift. Experimental results on synthetic data and the standard test videos show that in comparison with CABAC, the average PSNR is increased by 0.417 dB or the average bit rate is reduced by 2.11%.