Development of high efficiency image processing systems matched to the sampling structure of multidimensinal signal space

近年来，随着先进信息社会的发展，图像信息的数字化传输和积累变得越来越重要。作者已经提出了多维信号的采样定理，可用于图像的处理和编码。本课题针对低码率电视会议系统和静止图像数据库的高效图像编码方法进行了研究。(1)基于运动补偿和运动内插的图像编码方法本课题提出了一种新的低码率视频编码算法--带运动补偿和运动内插的图像编码方法。对于低码率视频编码系统，需要通过跳帧减少信息量，并对跳帧进行内插。传统的系统是在解码端对其进行内插，重建图像的质量是不够的。在该方法中，在编码端利用传输帧的位移向量对跳过的帧进行内插，并在内插后的图像信号和源信号之间传输误差信号。(2)基于内插的图像信号矢量量化提出了一种新的基于内插的矢量量化方法。该方法对SAB采样图像得到的原始图像和内插图像之间的误差信号进行矢量量化。结果表明，对于训练数据以外的图像，矢量量化器的性能恶化可以得到很大程度的改善。

The Object of the ProjectRecently, digital transmittion and accumulation of image information have gained importance following the growth of the advanced information society. The authors have already proposed the sampling theorem of multidimensional signal which can be developed for processing and coding of images. This project have aimed the research of highly efficient image coding method applied to a low-bit-rate TV conference system and a still picture database.The Results of the Project(1)Image Coding Scheme with Motion Compensation and Motion Interpolation In this project, a new low-bit-rate video coding algorithm which is named Image Coding Method with Motion Compensation and Motion Interpolation is proposed.It is necessary for the low-bit-rate video coding system to reduce information by skipping frames and to interpolate the skipped frames. The conventional systems interpolate them at the decoder, then the quality of the reconstructed images is not sufficient. In this method, the skipped frames are interpolated at the encoder by using the displacement vector of the transmitted frames and transmit the error signal between the interpolated image signal and source signal. It has been shown that this scheme improves the quality of the decoded images compared to the conventional method.(2)Vector Quantization Using Interpolation for Picture SignalsA new image coding scheme named Vector Quantization Using Interpolation is proposed. This method vector-quantizes the error signal between the original image and interpolated image made from sab-sampled image. It has been shown that the deterioration of the performance of the vector quantizer for images outside of the training data can be greatly reduced.