|
|
|
“Compression Artifact Removal and Inverse Halftoning Using Robust Nonlinear Filtering ”
by Mei-Yin Shen
May 1999
In this dissertation, a versatile nonlinear filtering approach based on the framework of robust estimation has been developed to solve two inverse problems: compression artifact removal and inverse halftoning. In the first part of this research, a low complexity postprocessing algorithm was proposed. We first formulate the artifact reduction problem as a robust estimation problem. Under this framework, the artifact-free image is obtained by minimizing a cost function that accounts for the smoothness constraint as well as image fidelity. Unlike the traditional approach that adopts gradient descent search for optimization, a set of nonlinear filters is used to calculate the approximate global minimum to reduce the computational complexity so that real-time postprocessing is possible. The proposed approach is generic and flexible. It can be applied to different compression schemes with minor variation. We have tested our algorithm on several standardized algorithms, including JPEG, JPEG 2000, and H.263, and demonstrated that our approach can alleviate compression artifacts efficiently with a low computational complexity. The elegant framework of robust estimation makes the proposed nonlinear filtering technique suitable not only for compression artifact reduction but also for other inverse problems such as inverse halftoning. Inverse halftoning plays a key role for allowing a wide range of operations to be performed on halftones. In the second part of this research, a blind inverse halftoning algorithm of low complexity was presented. The cascade of lowpass and nonlinear filters in conjunction with the edge enhancement technique produces high quality inverse halftoned images at a low computational cost. The proposed inverse halftoning algorithm outperforms existing techniques in terms of PSNR measurement and subjective quality. In addition, the proposed method is more robust w.r.t. different error diffusion kernels.