Anisotropic Diffusion as a Preprocessing Step for

Efficient Image Compression

Tamás Szirányi*, Ivan Kopilovic, Barnabás P. Tóth

University of Veszprém, Department of Image Processing & Neurocomputing

H-8200 Veszprém, Egyetem u. 10., Hungary

* Analogical and Neural Computing Laboratory, Comp. & Automation Inst., Hungarian Academy of Sciences, H-1111 Budapest, Kende u. 13-17, Hungary

14th ICPR, August 16-20 1998, Brisbane, IAPR, Australia

Keywords: Image compression, Anisotropic diffusion, Information content, Image enhancement

Abstract: Anisotropic diffusion was introduced in image processing as an image enhancement method. It is a non-linear smoothing process which intends to remove irrelevant or false details while preserving the edges, i.e. it "extracts" the essential visual information. The paper proposes a useful application of anisotropic diffusion in image data compression. We tried to show that for high compression a preliminary anisotropic diffusion on the image yields a better result after coding. This is an important achievement in view of low bit rate transmission.

Figure 1 The original image Clown

Figure 2 The result of the Perona-Malik diffusion

Figure 3 The original test image. Figure 4 Noisy image No.1. Gaussian noise, SNR= 10dB. Figure 5 Noisy image No.2. Salt and pepper noise, 15% affected pixels.

Table Compression results for the test image with and without preprocessing. Compression ratios indicated are the same for all images within the same rows.
No preprocessing (baseline JPEG) Preproc. with Perona-Malik diffusion Preproc. with CNN diffusion Preproc. with Pure an. diff. (Alv.-L.-Mor.)
Noisy image No.1 Compr. 9.7:1 (10.3%)

PSNR=14.34dB

PSNR=25.41dB

PSNR=20.13dB

PSNR=22.33dB

Noisy image No.2 Compr. 12.5:1 (8%)

PSNR=15.83dB

PSNR=25.03dB

PSNR=19.64dB

PSNR=23.83dB

Figure 6 Comparisons of PSNRs for compressions of Clown with and without preprocessing. Degree of compression is 100% for uncompressed images. PSNR is calculated versus the image indicated. AD denotes the anisotropic diffusion.

Figure 7 Detail of the encoded original (compression 7%)

Figure 8 Detail of the encoded Clown with Perona-M. diff. preprocessing (compression 7%)