Impact of JPEG2000 compression in correlation algorithms. Is subsampling a better compression strategy?

The main motivation of this paper is to analyze the impact of jpeg compression quantitatively. That is, how does compression affects the ability of measuring and comparing features in the image? A majority of processing and analyzing image techniques are based in comparing different pixel values, selected feature or entire patches in the image. These comparisons are even more important when dealing with image sequences or stereo pairs. When estimating motion or depth, the comparison of characteristic features or entire patches are of major importance. However, most studies interested in the evaluation of jpeg compression algorithms consider the visual quality of the compressed image as the unique criterium. Visual quality is very difficult to measure and remains a subjective evaluation. For example, in [5, 6, 7] the authors study the rate of compression acceptable for different medical imaging techniques. Their conclusion is that a compression of a factor 10 yields a good visual quality but superior rates of compression quickly degrades the image. Other studies like [9, 10] concentrate on technical aspects of the codification and transmission evaluating the simplicity, the completeness or transmission speed of different jpeg algorithms. In remote sensing a similar study to the one performed in medical imaging was performed in [12] dealing with stereo pairs. The conclusion was the same