Case Study On Image Compression And Its Integrated Circuit Implementation For Wireless Endoscopy System

As a new and promising system for digestive tract check, wireless endoscopy system has received increased interest these years. Low power and small size are two key specifications throughout the design of the capsule. Image compression is effective in reducing system power, transmitting bandwidth, and chip area of the capsule. Focusing on image capture and image processing, the main contributions of this dissertation are as follows.Firstly, two optimum datapath designs are proposed. By applying the datapath based on commercial image sensor and real-time data processing, the on-chip memory consumption and accessing are reduced by 2/3. The application of the datapath based on full custom Digital Pixel Sensor (DPS) can completely avoid extra on-chip memory further. Besides, by integrating image sensor with digital baseband, simple design and package, as well as more flexible readout sequence are available. Data overflow can be avoided, too.Secondly, two implementation-oriented lossless/near-lossless compression methods for BAYER image are proposed. The revised color channel splitting method improves compression ratio by filtering. Color channel splitting is realized with a local pixel conversion, the overhead of which is only 1.5 row memory. In the fully real-time processing method, a new prediction templet is presented to realize indirect data pattern conversion, so that the memory for buffering is avoided. According to a new bitstream accessing method, image data in different color channels can be processed in parallel in real-time. The complexity and hardware overhead of the fully real-time processing method are near to the JPEG-LS standard, while the compression rate is two times larger than that of JPEG-LS, under comparable reconstructed image quality.Thirdly, a VLSI (Very Large Scale Integration) implementation for the fully real-time processing method is proposed. In the compression engine, computations are arranged in a parallel pipelined architecture to realize real-time data processing. A specific clock management mechanism is implemented to ensure bottleneck calculations, as well as shuts off the working clock of idle modules, which reduces 15.7% of overall power dissipation. The proposed lossless/near-losslesss image encoder has been applied in wireless endoscopy system.Finally, a time domain DPS structured image sensor is proposed. By integrating the image sensor with other modules of wireless endoscopy system, the optimum datapath can be realized. Utilizing piecewise curve fitting in the proposed quantization method, 90% size of the quantization parameter table in uniform clock generator can be reduced. By applying self power gated technique, the average working time of the comparators is reduced to 6% of the image capture time. Based on the DPS image sensor and the compression engine, an on-chip camera is designed.