FORMULATION OF A 2-DIMENSIONAL MULTIPULSE MODEL AND ITS APPLICATION TO 2-CHANNEL CODING OF IMAGES

In recent years, discrete cosine transform coding is often applied as a highly efficiency coding method for images. In this method, however, quantization distortion, such as blocking effect or granular noise, a rises at low coding rates and the image deteriorates significantly. Th erefore, we propose a method where the above quantization distortion i s reduced by separating the image signal into the two components of th e high-frequency component and the low-frequency component based on th e principle of the two-channel coder, then each component is coded by a different method suited to the component's characteristics. That is, the low-frequency is coded by a high efficiency coding method. While the high-frequency component is coded based on the multipulse model be cause edge signals occur unpredictably in this component. This paper b egins by formulating the 2D multipulse model and derives an effective search method for 2D pulses, a 2D maximum correlation method. Next, ho w to accelerate this method is described and the number of computation s during pulse searches is dramatically decreased. Furthermore, how to apply the above method to the coding of images is explained. Finally, this method is applied to a standard image. We verify that blocking e ffects are suppressed at low coding rates and a reconstructed image ha ving distinct edges is obtained.