用自适应卡曼滤波处理活动图象

用自适应卡曼滤波处理活动图象 PROCESSING MOVING PCTUIRE BY USING ADAPTIVE KALMAN FILTER 王延平 Wang Yenping first-author 关于处理电视信号（如电视电话或会议电讯）的频带压缩或信息冗余度的压缩问题已有很多方法。【1-9】本文提出另一种处理方法。对于一个活动图象所需传送的信息仅是各帧信号之间的差一帧差信号,这一部分正好是图象的活动部分。由于人眼对于图象的活动部分的分辨力是不高的,所以我们可以传送帧差信号的近似值。为了得到帧差信号的线性近似,本文利用Kalman滤波。由于人眼在高亮度下可容许更大的亮度误差。因而Kalman滤波器被设计得在高亮度信号下比低亮度时容许更大的误差。这样在信号编码过程中可引入更大的容差,根据码率—容差理论,信息量可得到进一步压缩。为了对帧差信号的线性近似进行编码,我们只需对它们的端点进行编码,这样所传送的信息就仅为各端点的幅变和位置,进一步考虑到各行线段端点间的相关性,更多的信息冗余度可以被压缩。通过计算机模拟,证明了用自适应Kalman滤波线性近似来传送帧差讯号可大量压缩信息冗余度。例如,处理一个移动的手,对它图象编码的信息量约为0.3—0.4bit/pel. There are several methods to deal with the problem of redundancy reduction or compression of information for encoding a moving picture （eg. conference television or picture-phone）. In this paper auother method is presented. For a moving picture the information that has to be transmitted is the difference of the successive frames. Because of the poor resolution of human eye in the moving part of the picture, one can transmit an approximation of the difference signal. A Kalman filter is used to obtain a straight line approximation to the difference signal. Since the human eye is less sensitive to error in the intensity at higher value of this intensity, the Kalman filter is designed so as to emphasise intensity errors at lower values of the intensity. Therefore more tolerable distortion can be intruduced into the encoding procedure According to rate-distortion theory more bits can be saved. In order to encode the straight line approximation of the difference signal it is sufficient to encode its breakpoints. Consequently the information one has to sand is the positions and the amplitudes of the straight line approximation. By considering the dependace of the position and the amplitude of a breakpoint on a line with respect to those on the adjacent lines, more redundacies can be removed. Computer simulation has confirmed that straighe line approximation, using weighting factor that is a function of the signal intensity, can provide a different way to encode moveing pictures. Several sets of moveing pictures have been processed. As an example, to encode a moveing hand one needs about 0.3—0.4 bits per pel. 1980-04-01 2021-04-01 4