Co-histogram and its application in remote sensing image compression evaluation

Peak signal-to-noise ratio (PSNR) has found its application as an evaluation metric for image coding, but in many instances it provides an inaccurate representation of the image quality. The new tool proposed in this paper is called co-histogram, which is a statistic graph generated by counting the corresponding pixel pairs of two images. For image coding evaluation, the two images are the original image and a compressed and recovered image. The graph is a two-dimensional joint probability distribution of the two images. A co-histogram shows how the pixels are distributed among combinations of two image pixel values. By means of co-histogram, we can have a visual interpretation of PSNR, and the symmetry of a co-histogram is also significant for objective evaluation of remote sensing image compression. Our experiments with two SAR images and a TM image using DCT-based JPEG and wavelet-based SPIHT coding methods perform the importance of the co-histogram symmetry.