Restoring an image from its convolution with an unknown blur function is a well-known ill-posed problem in image processing. Many approaches have been proposed to solve the problem and they have shown to have good performance in identifying the blur function and restoring the original image. However, in actual implementation, various problems incurred due to the large data size and long computational time of these approaches are undesirable even with the current computing machines. In this paper, an efficient algorithm is proposed for blind image restoration based on the discrete periodic Radon transform (DPRT). With DPRT, the original two-dimensional blind image restoration problem is converted into one-dimensional ones, which greatly reduces the memory size and computational time required. Experimental results show that the resulting approach is faster in almost an order of magnitude as compared with the traditional approach, while the quality of the restored image is similar.

This paper describes the performance of the MPEG-4 still texture image codec in coding noisy images. As will be shown, when using the MPEG-4 still texture image codec to compress a noisy image, increasing the compression rate does not necessarily imply reducing the peak-signal-to-noise ratio (PSNR) of the decoded image. An optimal operating point having the highest PSNR can be obtained within the low bit rate region. Nevertheless, the visual quality of the decoded noisy image at this optimal operating point is greatly degraded by the so-called "cross" shape artifact. In this paper, we analyze the reason for the existence of the optimal operating point and the "cross" shape artifact when using the MPEG-4 still texture image codec to compress noisy images. We then propose an adaptive thresholding technique to remove the "cross" shape artifact of the decoded images. It requires only a slight modification to the quantization process of the traditional MPEG-4 encoder while the decoder remains unchanged. Finally, an analytical study is performed for the selection and validation of the threshold value used in the adaptive thresholding technique. It is shown that, the visual quality and PSNR of the decoded images are much improved by using the proposed technique comparing with the traditional MPEG-4 still texture image codec in coding noisy images.

We show that when using the MPEG-4 still image codec to compress a noisy image, increasing the compression ratio does not necessarily imply reducing the peak-signal-to-noise ratio (PSNR) of the decoded image. An optimal operating point is noted where the peak PSNR can be obtained within the low bit rate region. Yet, the visual quality of the decoded noisy images in this region is greatly degraded by the so-called "cross" shape artefact. To deal with this problem, we analyze the cause of the optimal operating point and the "cross" shape artefact when using the MPEG-4 still image codec for coding noisy images. We then propose an adaptive thresholding technique to remove the "cross" shape artefact. An analytical study is performed for the selection and validation of the threshold value used. It is shown that, by using the proposed technique, the visual quality and PSNR of the decoded images are much improved compared with the traditional MPEG-4 still image codec in coding noisy images.

We explore the wavelet coefficient selection and denoising by singularity detection (SD) for the embedded zero-tree wavelet (EZW) coding algorithm in this paper. The EZW coding algorithm exploits the relation between the multi-scale wavelet coefficients that finer scale wavelet coefficients are probably to vanish if the coarse scale wavelet coefficient vanishes. It is true for the parts of image that are regular but not the cases for noise-like features. In other words, the performance of the coding algorithm may be greatly degraded for the latter. In this paper, we investigate to arrange the wavelet coefficients according to the local regularity, by using the computed wavelet coefficients from the encoding filters. The advantage is two folds. For normal coding, we can make the encoded bit-stream first appear with the wavelet coefficients that correspond to the most regular part of the image, and the irregular one's follows. For noisy image encoding, we can remove noises before encoding hence increase the image quality as well as the coding efficiency.