Fast multi-resolution image processing for PCB manufacture

This paper addresses the problem of locating visual features on a printed circuit board (PCB) to facilitate automated manufacture and assembly of PCBs. A brief introduction to PCB manufacture is given to provide background information and demonstrate the large variation in types of PCB. An overview of visual servoing explains why the features must be located quickly and robustly. By presenting an algorithm which is able to locate features in the image of a PCB at any given scale quickly and then determine their position accurately, this paper aims to satisfy the criteria outlined. Following a discussion of the approach taken to detect features in the image, the theory behind a particularly efficient implementation is presented. Based on an approach to range searching in computational geometry, partial summation eliminates much of the processing which would otherwise be involved. It is particularly suited to multi-resolution image processing, and can be used both to find candidate feature locations and to confirm their precise position. Results which demonstrate the power of the technique for processing PCB images are given, and extensions to other applications are considered