Quality-controlled compression of sets of images

Abstract

We propose the QCLICS algorithm for quality-controlled JPEG compression of image sets. Lossy compression of images results in a tradeoff between compression and quality: the greater the compression, the poorer the resulting image quality. Quality-controlled compression has two goals: meeting any desired compression/quality specification accurately and optimizing the compression-quality tradeoff. When a large set of images is to be compressed into a limited amount of disk space, it is desirable to optimize the compression-quality tradeoff for individual images as well as across images. This is an important issue in multimedia CD-ROM production, where hundreds or thousands of digital images need to be compressed to fit in the available space. The QCLICS algorithm uses the RD-OPT algorithm (Ratnakar and Livny, 1995) for optimizing JPEG quantization tables for individual images, and uses a dynamic programming strategy to apportion the target compression/quality to the individual images. The complexity overhead of optimizing across the images is minimal, while significant gains in compression/quality are obtained. QCLICS offer flexibility in choosing the image quality metric and allows various constraints to be placed on compression/quality for individual images. We present performance results for several sets of images. These images were taken from various scientific sources at the University of Wisconsin-Madison, such as microscopy images of cells, crystals, brains, and spacecraft images of Venus.