In pattern recognition, data parallelization, the partitioning of an image into partial images and the processing of these images in parallel, is an effective and successful technique with respect to processing efficiency. However, since each partial image is processed independently, the consistency between partial images is not always assured. To solve this problem, it is desirable to directly exchange information among partial images. Thus, to attain a sufficient recogntion, ratio, we propose a pipeline processing mechanism for successive adjustment among neighboring partial images.