ANALYSIS OF TWO-STEP APPROACH FOR COMPRESSING TEXTURE IMAGES WITH DESIRED QUALITY

A task of lossy compression of remote sensing and other types of images with providing the desired quality is considered. Quality is mainly characterized by the peak signal-to-noise ratio (PSNR) but visual quality metrics are briefly studied as well. Potentially, a two-step approach can be used to carry out a compression with providing the desired quality in a quite simple way and with a reduced compression time. However, the two-step approach can run into problems for PSNR metric under conditions that a required PSNR is quite small (about 30 dB). These problems mainly deal with the accuracy of providing a desired quality at the second step. The paper analyzes the reasons why this happens. For this purpose, a set of nine test images of different complexity is analyzed first. Then, the use of the two-step approach is studied for a wide set of complex structure texture test images. The corresponding test experiments are carried out for several values of the desired PSNR. The obtained results show that the two-step approach has limitations in the cases when complex texture images have to be compressed with providing relatively low values of the desired PSNR. The main reason is that the rate-distortion dependence is nonlinear while linear approximation is applied at the second step. To get around the aforementioned shortcomings, a simple but efficient solution is proposed based on the performed analysis. It is shown that, due to the proposed modification, the application range of the two-step method of lossy compression has become considerably wider and it covers PSNR values that are commonly required in practice. The experiments are performed for a typical image encoder AGU based on discrete cosine transform (DCT) but it can be expected that the proposed approach is applicable for other DCT-based image compression techniques.