Ultra-High-Speed Real-Time Processing Circuit and Algorithm

为了实现实时有效的图像压缩，我们提出了自适应分辨率矢量量化技术（AR-VQ）。自适应分辨率转换方法根据像素块纹理的形式自适应地改变图像的分辨率。对于包括文本的图像，AR-VQ在压缩图像质量方面实现了比世界标准JPEG和JPEG 2000高得多的上级压缩性能差，为5至40 dB。我们还开发了一种先进的矢量量化编码方法，并已实现到硬件架构。该处理器采用了不需要计算的消除方法和AR-VQ技术。为了验证采用该处理器的静止图像编码系统的性能，开发了一个评估板。AR-VQ算法和无线网络技术使无线局域网数据投影仪的研制成为可能。另外，为了解决实时图像识别的问题，我们提出了一种简单而可靠的基于VQ直方图的人脸识别方法。相邻像素灰度差量化直方图方法实现了视频速率下的人脸识别。实验结果表明，该方法对40人400幅图像（每人10幅图像）的识别时间在31 msec以内，识别率达到95.7%，与VQ直方图法相当，并提出了一种变异性提取方法，以实现高可靠性的LSI设计。该方法可以互相测量芯片间变异性和芯片内变异性。

In order to realize real-time efficient image compression, we have developed the adaptive resolution vector quantization technique (AR-VQ). The adaptive resolution conversion method changes the resolution of image adaptively according to the form of a pixel block texture. The AR-VQ has realized a much superior compression performance difference of 5 to 40 dB in compressed image quality than the worldwide standard the JPEG and the JPEG2000 for the images including text. We have also developed an advanced VQ encoding method and have implemented it into hardware architecture. The processor employs the needless calculation elimination method and the AR-VQ technique. An evaluation board has been developed in order to verify the performance of still-image encoding system employing the processor. The AR-VQ algorithm and wireless network technology make it possible to produce a wireless LAN data projector, in 2002.In addition, we have developed a very simple yet highly reliable face recognition method based on the VQ histogram method in order to resolve real-time image recognition. The adjacent pixel intensity difference quantization histogram method has realized human face recognition at video-rate. Experimental results show a recognition time for 400 images of 40 persons (10 images per person) within 31msec, maintaining a recognition rate of 95.7% equal to the VQ histogram method.Furthermore, we have developed an extracting methodology of variability in order to realize a high-reliability LSI design. The method can measure inter-chip variability and intra-chip variability each other.