科研之友
微信
新浪微博
Facebook
分享链接
摘要
In this paper, we present a new de-blocking algorithm for highly compressed images by soft-thresholding in wavelet domain. The main point here is to improve the deblocking effect with better-estimated threshold values. This is based on two observations. First, the non-texture areas of compressed images can be restored accurately with low pass filtering, and therefore, provide better reference in estimating thresholds. Secondly, wavelet transform provides good approximation at coarse scale, which is actually the effective low pass filtering. Accordingly, we propose two approaches to estimate thresholds. Both approaches use proper references in the estimation and obtain thresholds for all high sub-bands directly, while the computational complexities are comparable to or even lower than the known best algorithm. Experimental results prove the performance of the new algorithm both subjectively and objectively. Moreover, the convergence of the estimated thresholds not only demonstrates the reasonableness of the new algorithm, but also indicates a very important feature of wavelet-based soft-thresholding, namely, the quantization parameter, e.g. quantization table, is the dominant factor to determine the thresholds and the visual deblocking effect. This is of practical significance for real time applications.
