Filtering operation method for a very high-speed image processing system.

摘要

In an image processing system for performing parallel processing for each pixel by line parallel local computation an endless type slit memory (4) of finite length capable of storing as much data as required in a row, in which a terminal address is connected to a top address, is addressed by a base pointer (1) and a reference pointer (2). Each time the line is changed, the content of the base pointer is changed one by one by an external controller (50), and it is cleared to a top address after a terminal address, the content of the base pointer is written in the reference point at the same time, and the address of the slit memory (4) is specified by the reference pointer (2) and the content of the reference pointer is changed by one at the same time, and a region of the slit memory corresponding to the mask size of a filter is sequentially accessed by the reference pointer. In a novel filtering method for such system using a filter having weight with left-to-right symmetry, each processing element (PE) multiplies the pixel data of a row, for which it is in charge of processing, by the weight of the central row of a filter and the results are added, and the pixel data of a row, for which it is in charge of processing, is multiplied by the weight of the other rows on one side of the central row of the filter and the results are added. Each element receives the results of computation maintained by the processing elements of the other rows by transfer and executes a filtering operation.