2次元離散ウェーブレット変換のハードウェアアーキテクチャとセミフラジャイル電子透かしシステムへの応用に関する研究

摘要

This thesis describes a study done on a 2-dimensional wavelet transform hardwarearchitecture that has high throughput, a small memory area, and low energyconsumption and an evaluation of it. Also, the thesis describes an applicationstudy done to evaluate the effectiveness of the wavelet architecture. For the applicationstudy, we used an architecture of a novel digital watermark that detectstampering in a digital image. The method has low overhead in countering illegalattacks to copy digital watermarks. We describe the design of the hardwarearchitecture and our evaluation of it with the wavelet architecture.A wavelet transform has been used for various applications such as image/videocompression, recognition, and watermarking because of its useful characteristicssuch as time-frequency analysis. Also, demand has increased for mobile devices tohandle the many complex applications that can be done on personal computers.In these mobile devices, applications are implemented into the hardware becausethey should have low energy consumption with high throughput. As a result, awavelet architecture that has high throughput, small memory, and low energyconsumption is required. However, no one has been able to design a waveletarchitecture with high throughput, small memory, and low energy consumptionyet.We focused on optimizing memory access by designing two architectures toeliminate the bottleneck of the wavelet operation. One of the architectures balancesthroughput, memory area and energy consumption. The other optimizeshigh throughput and low energy consumption. The evaluation results show thatthe throughput, energy consumption, and memory area of the balanced architecture are 10 times higher, 10 to 20 times lower, and 4 to 50 times lower thanthat of conventional architectures, respectively. Furthermore, the throughput ofthe optimized architecture for high throughput and low energy consumption is20 times higher than, the energy consumption is 20 to 30 times lower than, andthe memory area is comparable with that of the conventional architectures.We applied the two aforementioned architectures to a watermarking applicationto evaluate the effectiveness of the wavelet architecture. Watermarking isone of the information hiding technology for music, image, and video for copyrightauthentication and tampering detection. This thesis describes a new semifragilewatermarking that detects tampering. We also designed an watermarkinghardware architecture for mobile devices. Our watermarking architecture wascombined with the aforementioned wavelet architectures, and its performancewas evaluated.The results show that the throughput and energy consumption of the conventionalwavelet architecture is lower and higher than those of the watermarkingarchitecture, while those of our wavelet architectures are comparable to thoseof the watermarking architecture. Also the size of the internal memory of ourarchitectures is lower than that of the watermarking architecture. These resultsare due to the fact that the 2-D DWT dataflow is optimized for DRAM accessand that coefficients that should be stored in the internal memory are reduced.These results provide evidence that our wavelet architecture has high throughput,a small memory area, and low energy consumption. The effectiveness of thearchitecture was also demonstrated in our application study.