Pattern association from noisy images by the network constraint analysis

This paper describes a technique for realizing visual association by the network constraint analysis. In order to make machine visual processing more practical and powerful, it is important to develop a technique for understanding various noise superimposed images. To solve this problem, a two-stage association technique is proposed based on network constraint analysis. In the first stage, strict screening of the memory which contains reference images is performed employing an acquired noisy image to yield an intermediate image, while in the second stage, rather weaker screening of the memory is done by the intermediate image and depth first search is applied to those surviving image pieces in the memory to finally obtain an associated image. An algorithm to speed up the association process is also employed in the second stage. Performance of the two-stage association is examined by an experiment employing real noisy alphabetical images and satisfactory results are obtained