Despite several more advanced image compression algorithms have been proposed, JPEG is still widely used because of its advantage in simplicity. However, images encoded by the JPEG algorithm exhibit undesirable blocking and ringing artifacts. In particular, for document images, ringing artifacts reduce the shapeness and clarity of the text, and affect the readability of the documents. This dissertation presents two different approaches to improve the decoding quality for document images encoded with JPEG.;In the first approach, we pose the JPEG decoding problem as an inverse problem and decode the image with Bayesian reconstruction. The scheme works by first segmenting the image into blocks of three classes corresponding to background, text, and picture. For each class of image blocks, we design a specific prior model to capture the characteristics of the class. The class-specific prior models are then used to compute the maximum a posteriori (MAP) estimate of the original image. The scheme substantially improves the quality of decoded images both visually and as measured by PSNR. Also, the decoded text regions are essentially free from ringing artifacts even when the images are compressed at a low bit rate.;In the second approach, we introduce the Hypothesis Selection Filter (HSF) as a generic approach for image quality enhancement. The HSF provides a systematic method for combining the advantages of multiple linear or nonlinear image filters into a single general framework. Our major contributions include the basic architecture of the HSF and a novel unsupervised training procedure for the design of an optimal pixel classifier. The resulting classifier distinguishes the different types of image content and appropriately adjusts the weighting factors of the image filters so that each filter is applied to the regions for which it is most appropriate. We demonstrate the effectiveness of the HSF by applying it as a post-processing step for JPEG decoding so as to reduce the JPEG artifacts in the decoded document image. In our scheme, we incorporated 4 different image filters for reducing the JPEG artifacts in the different types of image content that are common in document images, like text, graphics, and natural images. Based on several evaluation methods, including visual inspection of a variety of image patches with different types of content, global PSNR, and a global blockiness measure, our method outperforms state-of-the-art JPEG decoding methods. In addition, the generic structure and the training basis of HSF makes the scheme potentially applicable to many other quality enhancement and image reconstruction tasks.
展开▼
