Lossless image compression method using vector quantization based on minimizing mean absolute error

Abstract

In this paper, we propose a novel lossless image compression method. During the prediction stage for each block of 8×8 pixels, a mechanism for preselecting one of �� linear predictors from the dictionary is employed. The dictionary is determined individually for each encoded image using vector quantization (initially with a redundant number of vectors in the dictionary) and a fast algorithm that minimizes mean absolute error. In next steps, the prediction errors are encoded in a two-step manner using an adaptive Golomb code followed by an adaptive binary arithmetic coder. In this study, we demonstrate the efficiency of the proposed solution against other competitive codecs, including those based on deep learning. The proposed method offers high compression efficiency and is characterized by a short decoding time.