Quantization of color image components in the DCT domain

Abstract

Many standard image compression techniques involve applying discrete cosine transforms (DCTs) to the components comprising a color image and quantizing the resulting transform coefficients. Such techniques can be applied to color images in any color coordinate space. To achieve maximum compression, the DCT coefficients should he quantized as coarsely as possible, while allowing no visible distortion in the decompressed image. Due to variations in the contrast sensitivity of the human visual system as a function of spatial frequency, different DCT coefficients require different precision in their quantization in order to meet this goal. Different color components may also have different quantization requirements. In this paper we propose a DCT coefficient quantization scheme for color images in the RGB color space. We will describe experiments performed to determine human visual system sensitivity to the DCT basis functions in the RGB space, and how the results from these experiments are used to design the quantization scheme. We will then compare our quantization scheme to other techniques proposed for images in the RGB and YCrCb color spaces, in terms of quality of the decompressed images, and data compression achieved.