Analysis of Impact of Lossy Image Compression on Image Quality: From Image Application Aspects

Abstract

Reliable methods for evaluating image compression quality are needed to assess the effects caused by those algorithms, and such assessments require an application-driven process. This paper aims to provide a comprehensive analysis of the impacts of JPEG2000 compression on remote sensing image applications. Image classification and image matching quality assessment, which indicate the radiometric and geometric quality of the compressed images, were selected to test the compression impacts. Nine remote sensing images with different terrain types, land cover and image resolution were employed as original images for compression. The images were compressed by JPEG2000 at various compression ratios. In image classification quality assessment, the classification result from the original image was regarded as the ground truth. The results from original images and compressed images were then compared. In image matching accuracy assessment, the ‘pixel distance’ between the original image and the compressed image was computed to indicate the geometric distortion caused by image compression. From the experiments, the image quality would decrease with the increase in the compression ratio. As a result, the accuracy of image classification and the matching accuracy deteriorate when the image JPEG2000 compression ratio is very high. And the effect of JPEG2000 compression on the accuracy of image classification and image matching is generally linear.