Novel Learning- and Texture-Based Approach with Applications to Image Coding and Inverse Halftoning
| Autor: | Yong-Huai Huang, 黃詠淮 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 Texture is an important feature in images and has been widely used in many applications. Based on the classified textures, this thesis presents a novel learning–and texture–based approach to design more efficient image and video processing algorithms such as the adaptive arithmetic coding for error–diffused images, the predictive coding for gray images, and the inverse halftoning algorithm to reconstruct gray images from halftones. For context–based arithmetic coding, the block– and texture–based training process is first applied to train the multiple–template according to the most representative texture features. Based on the trained multiple–template, we next present a texture– and multiple–template–based (TM–based) arithmetic coding algorithm for lossless compression of error–diffused images. In our proposed TM–based algorithm, the input image is divided into many blocks and for each block, the best template is adaptively selected from the multiple–template based on the texture feature of that block. Experimental results demonstrate that the compression ratio of our proposed TM–based algorithm is superior to that of joint bilevel image group (JBIG) standard and the previous algorithms proposed by Reavy and Boncelet and by Lee and Park. For predictive coding, to improve the accuracy of the least square–based prediction scheme, we present a new texture–based training process to construct the multiple–window for various image contents. Based on the trained multiple–window, the texture–and multiple–window–based (TMW–based) prediction scheme is presented for lossless compression of images. In our proposed TMW–based scheme, for each pixel of the input image, the best training window is adaptively selected from the multiple–window according to the texture feature. Experimental results demonstrate that the accuracy of our proposed TMW–based prediction scheme is better than that of the previous LS–based prediction scheme. For inverse halftoning, based on our proposed variance gain–based DT (VDT), a texture and VDT (TVDT)–based training process is presented to construct a lookup tree–table which will be used in the reconstructing process. In the reconstructing process, to enhance the quality of the non–smooth regions in the reconstructed image, we propose an edge–based refinement scheme to enhance the quality of the reconstructed gray image. Experimental results demonstrate that our proposed TVDT–based inverse halftoning algorithm has the highest image quality when compared to the currently published three inverse halftoning algorithms, such as the DT–based algorithm, the lookup table–based algorithm, and the edge–and lookup table–based algorithm. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|