A method for enhancing the picture quality of low bit-rate block-coded images

• Published in: Fifth International Conference on Image Processing and its Applications pp. 40 - 44
• Main Authors: Sampson, D.G, Papadimitriou, D.V, Ghanbari, M, Dennis, T.J
• Format: Conference Proceeding
• Language: English
• Published: London IEE 1995
• Subjects: Codes; Optical information, image and video signal processing; Pattern recognition; image classification; picture impairments removal; staircase effect; post-filtering algorithm; smoothing methods; low bit-rate block-coded images; filtering process; Gaussian smoothing operator; adaptive signal processing; image enhancement; post-processing method; low detail image regions; quantization noise; block truncation coding; adaptive filters; spatial domain block coding techniques; Gaussian kernel; smoothed image gradient data; blocking effect; high detail image areas; picture quality enhancement; vector quantisation; region classification; simulation results; vector quantization; adaptive Gaussian-type filter; image coding
• ISBN: 0852966423; 9780852966426
• DOI: 10.1049/cp:19950616

We describe a post-processing method which is designed to remove the picture impairments typically introduced by spatial domain block coding techniques, such as vector quantization (VQ) and block truncation coding (BTC). In these coding methods, the quantization noise is highly correlated with the characteristics of the original signals, so that different areas of the coded image suffer from distinctly different impairments. In particular, uniform areas suffer from the blocking effect, where the boundaries of adjacent blocks become visible resulting in false contours. Moreover, all diagonal edges appear jagged, which is known as the staircase effect. Finally, most of the textured areas are `washed-out'. The post-processing method exploits the properties of smoothed image gradient data to distinguish between high and low detail regions of the decoded image. The filtering process is then decided based on the region classification. For the high detail image areas, which include edges and texture, a fully adaptive Gaussian-type filter is employed. The shape and the position of the Gaussian kernel is adapted according to the characteristics of each local region. For the low detail image regions, which include areas of constant or slowly varying intensity, a simple Gaussian smoothing operator is used. Simulation results confirm that, after applying this post-filtering algorithm to the coded images, the annoying blocking effect is reduced and the jagged edges appear continuous and sharp.