Prediction of Visual Quality for Lossy Compressed Images

Compression is a typical operation applied in image processing. Lossy compression is widely used although it introduces distortions that might lead to degradation of image visual quality. Then, introduced distortions have to be controlled at compression stage to provide visually lossless compression or appropriate quality according to adequate metrics. Often this should be done without iterations of image compression/decompression. In this paper, we propose an approach to predict visual quality metrics that, for coders based on discrete cosine transform, depend on quantization step or scaling factor. Using such a dependence, it is possible to set quantization step or scaling factor properly (with high accuracy) before carrying out lossy compression. Experiments are performed for optical and remote sensing test images showing that the proposed approach is quite universal.