Optimization of quantization table based on visual characteristics in DCT image coding

Abstract

The DCT (Discrete Cosine Transform) based coding process of full color images is standardized by the JPEG (Joint Photographic Expert Group). The JPEG method is applied widely, for example a color facsimile. The quantization table in the JPEG coding influences image quality. However, detailed research is not accomplished sufficiently about a quantization table. Therefore, we study the relations between quantization table and image quality. We examine first the influence to image quality given by quantization table. Quantization table is grouped into four bands by frequency. When each value of bands is changed, the merit and demerit of color image are examined. At the present time, we analyze the deterioration component of a color image. We study the relationship between the quantization table and the restoration image. Color image is composed of continuoustone level and we evaluate the deterioration component visually. We also analyze it numerically. An analysis method using the 2-D FFT (Fast Fourier Transform) can catch a change of a color image data by a quantization table change. On the basis of these results, we propose a quantization table using Fibonacci numbers.