自适应图像去振铃效应滤波器

Image Deringing with Adaptive Bilateral Filter1

Zhai Guangtao, Xu Yi*, Yang Xiaokang, Zhang Wenjun, Yu Songyu Institute of Image Communication and Information Processing, Shanghai Jiao Tong University,

Shanghai, China (200240)

Abstract

In this paper, we tailor the bilateral filter towards the task of suppressing the ringing artifact commonly occurred on JPEG2000 images under low bitrates. The proposed adaptive bilateral filter varies from its original form as the pixel moves from a monotone area towards an edge one. Also the local spreads of the domain and range filters are tuned with the extent of texture activity. The edge detection and distance transform are used to indicate the local edge and texture activity indexes. Experimental results show that the adaptive bilateral filter can effectively smooth out the annoying ringing artifact and ameliorate the visual quality.

Keywords：Bilateral Filter, Image Postprocessing, Deringing

1.Introduction

Ringing is a kind of Gibbs phenomenon, which is caused by heavy truncation on transform coefficients and manifests itself as spurious oscillations around strong edges. Also, ringing can come from improper image restoration operations [1]. The ringing artifact encountered in the new image coding standard JPEG2000 is much more difficult to model and/or suppress than the blockiness artifact in the last generation block-based coding standard (JPEG). Among the deringing algorithms, the postfiltering schemes are the most attractive due to their compatibility with existing standards and codecs. Based on O’Rourke and Stevenson’s work on blockiness reduction [2], Shen and Kuo [3] formulated deringing into a classical maximum a posteriori (MAP) estimation problem using a Markov random field (MRF) model, and further proposed a non-iterative nonlinear filter to approximate the global optimum solution. Oguz et al. [4] proposed to use combined binary and grayscale morphological operations to filter out ringing artifact. And they also suggested a new perceptual ringing artifact measure named visible ringing measure (VRM) [5]. Fan and Cham [6] designed an edge model under the framework of multiscale edge analysis and used it to reconstruct the corrupted edges in low bitrate wavelet coded image. Nosratinia [7] re-applied JPEG2000 compression on a redundant representation of pixel-by-pixel shifted images and finally integrated the shift-backs to generate the postfiltered image. Essentially, this approach is thought to be deeply related to translation-invariant denoising algorithms introduced in [8]. Yang et al. [9] employed a maximum likelihood estimation approach together with a k-means algorithm and a cluster-segmentation processing to suppress ringing artifact. Recently, Tan and Wu [10] designed a vision model for postfiltering JPEG2000 coded color images. Their model considers both inter and intra band visual masking effects to guarantee a HVS plausible processing result. This postfiltering algorithm, however, is designed for a specific codec designed by the authors themselves [11], and this somewhat restricted its usage. Chen et al. [12] applied grayscale morphological operation together with a voting stage to choose an optimal postfiltering for deringing JPEG2000 images on the encoder side. Consequently this algorithm needs extra bitrate overhead of the morphological filter details to be transmitted to the decoder, and thus is not compatible with the existing standards. And more

1 This work was supported by National Natural Science Foundation of China (60332030, 60502034, 60625103,60703044), Shanghai Rising-Star Program (05QMX1435), Hi-Tech Research and Development Program of China 863 (2006AA01Z124), NCET-06-0409, the 111 Project and the specialized Research Fund for the Doctoral Program of Higher Education under grant No. 20040248047.