應用廣義重要位元平面轉移之可調適JPEG2000感興趣區域編碼技術之研究

Adaptive Generalized Significant Bitplanes Shift Method for JPEG2000 Region of Interest Image Coding

黃建霖(Chien-Lin Huang)

指導教授 ： 郭文嘉

元智大學/資訊學院/資訊管理學系/碩士(2005年)

摘要

JPEG2000影像壓縮標準中對於感興趣區域(region of interest, ROI)編碼方式的支援是一項創新且極具意義的功能，其主要的ROI編碼方式有最大位移法(Maxshift method)及一般位移法(general scaling based method) 。最大位移法提供一個在固定位元率之下使得影像中感興趣區域獲得該位元率下最佳影像品質的編碼方式，而一般位移法則提供一個強調感興趣區域影像品質、同時兼顧背景資訊的編碼方式。本論文主要針對JPEG2000編碼方式提出一個新的ROI位元平面轉移分配方式，除了保留影像中ROI區域和背景區域的相對品質可調整的特性外，並提供一個滿足使用者ROI影像品質要求下，尋得最佳背景品質的編碼方式。除此之外，我們進一步利用貝氏信任網路(Bayesian Belief Network)以影像特性、使用者對ROI的品質需求及對影像的壓縮位元率等特徵值作為選取最適合的ROI位元平面轉移參數作為編碼時的依據，以提供JPEG2000影像壓縮標準中對於ROI位元平面轉移參數選取的模型。實驗結果顯示，本論文所提之方法可以有效預測符合使用者要求品質的編碼方式，有效增進ROI影像編碼的效率及彈性。

關鍵字

jpeg2000 ； 貝氏網路 ； 位元率 ； 位移法 ； 使用者 ； 特徵值 ； 影像

並列摘要

The support of region of interest (ROI) coding is an innovative feature of JPEG2000 image coding standard. It defines two kinds of ROI coding methods—the maximum shift (Maxshift) method and the general scaling based method. The former allows arbitrarily shaped ROI coding such that ROI can acquire best quality under fixed bit ratio without explicitly coding of shape information while sacrifices the flexibility of adjusting scaling value for determining the degree of relative importance between ROI and background wavelet coefficients. The latter allows a rough control on both ROI and background qualities distributions in the coding stream. In this study, we propose a novel ROI-based coding method that extends the scaling-based method of JPEG2000 as well as combines the advantages of the two ROI coding methods defined in JPEG2000 standard and other existing ROI coding methods. It also provides an explicit way with Bayesian network for helping user to choose suitable bitplane shift parameters according to the image properties, the user’s ROI quality requirement and the compression bit ratio for encoding ROI in JPEG2000 images. The proposed method mainly provides some hints on how to choose an optimal scaling based value and how to use the ROI extra bits to improve the significant quality of the background during the shift operation. Experimental results demonstrate that the proposed method can effectively forecast the encoding procedure that meets the requirement and increase the efficiency and flexibility of ROI image encoding.

並列關鍵字

region of interest ； jpeg2000 ； image feature ； Bayesian Belief Network