Artifact reduction with diffusion preprocessing for image compression

Tamás Szirányi Hungarian Academy of Sciences Analogical and Neural Computing Laboratory Computer and Automation Instrumentation H-1111 Budapest Kende u. 13-17 Hungary and University of Veszprém Department of Image Processing and Neurocomputing H-8200 Veszprém Egyetem u. 10, Hungary E-mail: [email protected] Abstract. We evaluate a preprocessing method for image compression artifact reduction based on nonlinear diffusion filtering that we proposed earlier. The method consists of using edge-adaptive diffusion processes before the discrete cosine transform (DCT)-JPEG compression. By using a simple measure for artifact reduction, we show that considerable artifact reduction is achieved with preprocessing at the same bit rate as, and with no greater error than, the original compression. We also show that preprocessing helps to preserve the true contours for image processing applications. An automatic parameter selection for the preprocessing is also proposed, considering the edge histogram of the image and depending on the compression ratio. We test the method for visual quality with extensive subjective measurements. We show that, depending on the image content, preprocessing can significantly improve the visual quality at low bit rates. © 2005 Society of Photo-Optical Instrumentation Engineers. [DOI: 10.1117/1.1849242]