Quantization for predictor coefficients on predictive coding

摘要

The predictive coding strategies employ a linear predictor in general. The predictor coefficients are obtained by solving the normal equations in minimization of the total squared prediction residuals. The coefficients are decimals; however the values, for example 1/2, 1/4, and 3/4, are convenient since the predicted values are computed with shift operation, addition, and/or subtraction. In this paper we show simple predictors as follows: (1) the predicted valises are obtained by weighted summation of the value immediately above and immediately to the left of the current pixel. (2) the predictor coefficients sum up to one. The differences of the total squared residuals between the simple predictors and the predictors obtained by solving normal equations are discussed in a theoretical approach. The estimations for actual images are also described.