A JPEG-LS Based Region of Interest Coding for Lossless Compression of Capsule Endoscope Images

• Main Authors: Ching-Hung Chu, 朱鏡宏
• Other Authors: Shaou-Gang Miaou
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/03623988168376204308

碩士 === 中原大學 === 電子工程研究所 === 95 === Abstract As the development in digital technology continues to get progress rapidly, the digital data sizes of medical images increase quite enormously. In the application of telemedicine, transmission of such enormous amount of medical data becomes a heavy loading via a bandwidth-limited network. Similarly, in the archiving application, the requirement for saving such an enormous amount of data to limited storage space is a heavy burden too. Capsule endoscope is a typical source to create this kind of problem. One of the useful solutions to this problem is data compression. Furthermore, to aviod medical dispute, lossless compression is a better choice to compress capsule endoscope images than a lossy one. The thesis intends to propose a high-performance compression method that is particularly suitable for lossless compression of capsule endoscope images. In this thesis, a JPEG-LS-compatible compression algorithm based on a special padding scheme is presented to handle the images with arbitrary shapes and sizes of ROIs (region of interest) in a fixed template format. The proposed method is applied to the lossless compression of capsule endoscope images with diagnostic information contained in ROI zones. We use JPEG-LS which had already been proved with its excellen coding performance and low complexity feature in static image compression. With well-established JPEG-LS technology for image compression, it is needless to develop a whole new compression algorithm. All we need is to exploit the ROI characteristics of capsule endoscope images and be aware of the prediction requirement of JPEG-LS, and focus on the design of a good interface to the original technology. Experimental results show that this ROI-based method achieves better compression performance than a full-sized image based JPEG-LS method does. Specifically, the proposed method achieves 17% improvement in compression ratio, and saves the encoding time by about 59%. In the field of lossless compression, these results can be valuable and commendable. The proposed method in this thesis is also suitable for other ROI-based medical image compression applications, such as mammography, computed tomography, etc.