Bit-Error Aware Lossless Image Compression

In this paper, we propose and investigate simple bit-error aware lossless compression algorithms for the compression and transmission of image data under the bit-error environment. We focus on enhancing two-stage lossless compression algorithms. The first stage uses a simple linear predictor, whereas at the second stage, we apply bi-level block coding, interval entropy coding, and standard entropy coding. The key coding parameters of the predictor, bi-level block coding, or entropy coding parameters are protected by the usage of a forward error correction scheme such as (7,4) Hamming coding. The residues from bilevel bloc coding or the residue offsets from Huffman coding are not protected to compromise the performance of compression ratio. Our compression experiments demonstrate that when the bit error rate (BER) in the channel is equal to or less than 0.001, the lossless compressed image can be recovered with a good quality. Introduction Lossless image compression methods are usually required for compression and transmission of image data whenever a lossy compression approach cannot be applied. Such systems include medical imaging, remote sensing, and high cost archiving systems. Especially in the medical data transmitting and archiving system, the usage of lossy compressed images for diagnostic purposes is prohibited by law in many countries. It is also preferred for the image data of mechanical fault diagnosis to be lossless compressed. In general, a lossless compression algorithm consists of two stages as described in references [1-6]. The first stage performs predictions to remove data correlation in order to produce residue data. The resultant residues have reduced amplitudes and are assumed to be statistically independent with an approximate Laplacian distribution [1-3]. The second stage further compresses residue data using an entropy coding algorithm—that is, Huffman coding or arithmetic coding [1-6]. Much research work has been conducted to improve the compression ratio using a more complex predictor as well as either adaptive Huffman or arithmetic coding with increased algorithm complexity. In addition, the usage of lossless image compression could improve transmission throughput if the compressed image data is transmitted over a network system. However, if bit errors occur in a noisy channel during transmission or in the storage Proceedings of The 2011 IAJC-ASEE International Conference ISBN 978-1-60643-379-9 media, the recovered image will be damaged and will become useless. This outcome results from the fact that a standard entropy coder generates instantaneous codes, which are sensitive to bit errors. Although this problem can be cured by applying a forward error control scheme [7], adding additional bits required by the error correction coding can significantly degrade the performance of the compression ratio and may even cause the expansion of image files. For example, if an 8-bit grayscale image is lossless compressed to 5 bits per pixel, using a (7,4) Hamming code (adding three parity bits for every 4 data bits for a single bit error correction) for bit-error protection will increase the compressed data size by 75%; that is, 8.75 bits per pixel, which indicates image file expansion. This paper investigates two new simple algorithms: predictive bi-level block coding and predictive interval entropy coding for bit-error aware lossless image compression. To gain a compromised compression ratio, only the prediction parameter, bi-level block coding and interval entropy coding parameters are protected by the (7,4) Hamming codes. The standard predictive entropy coding with bit error correction is also included for comparison purposes. We evaluate and compare the algorithm performances in terms of the compression ratio and the peak signal to noise ratio (PSNR) versus the bit error rate (BER). Bit-Error Aware Two-Stage Lossless Image Compression Figure 1 shows a block diagram of our bit-error aware two-stage lossless compression algorithms. For the first predictive stage, we use up to three neighboring pixels as depicted in Figure 2: the left-hand neighbor (A), the upper neighbor (B), and the upper-left neighbor (C). X is the predicted pixel. Predictor Residue coding Residues Image data Predictor parameters, residue coding parameters protected by Hamming coding (7,4) Packing scheme Bit stream Figure 1: Bit-error aware two-stage lossless image compression scheme