Parameterization of the nonlinear Greham predictor for digital image compression

Parameterization of the nonlinear Greham predictor is performed for a digital image compression method based on the differential pulse code modulation. The predictor automatically selects different methods for calculating each pixel in the image based on the availability and intensity of that pixel contour. A fast learning procedure that optimizes the prediction parameters is performed before the actual compression. In the course of optimization, the minimum sum of absolute values of prediction errors is provided. For this purpose, a recursive procedure is used, whose computational complexity is independent of the image size. The estimation of the computational complexity of the proposed predictor is conducted. To study the predictors computational experiments are carried out on real images. A gain the proposed predictor offers in terms of the root mean square error when compared with the prototypes is demonstrated. In addition, a gain that the compression method based on the differential pulse code modulation with the proposed predictor has over the JPEG compression method in terms of the maximum error is demonstrated.