Lossless image compression using BPNN predictor with contextual error feedback

In this paper, a lossless image coding scheme is proposed, which is based on Back Propagation Neural Network (BPNN) pixel value predictor, contextual error feedback and context adaptive arithmetic coding. The use of BPNN predictor in the proposed image compression scheme has resulted in improved compression ratio in comparison to the benchmark lossless compression schemes, LOCO-I and CALIC. The main advantage of BPNN pixel value predictor is its capability to adapt itself to the image being coded. The use of BPNN predictor resulted in lower values for entropy as compared to other predictors such as, MED (Median Edge Detection predictor) and GAP (Gradient-Adjusted Predictor) used in LOCO-I and CALIC respectively. The average bits per pixel (bpp) value obtained by utilizing the proposed scheme for nine test images indicate an improvement by 7.6% over LOCO-I, 2.85% over CALIC and 8.82% over JPEG2000 in lossless mode.