Gray-scale vision system based on multiple cell-feature extraction engine

We presents a flexible and highly-reliable gray-scale vision system besed on multiple cell-feature extracting engine composed of multi-layer feature-plane and two basic operation modules: extended convolution and radially traversing probing. This engine can easily extract various image features such as moment, edge curvature, complexity, extent, blobs, and bars. We define“multiple cell-features”as a multi-dimensional feature which comprehensively represents properties of a subimage, which is here named a“cell”. The generalized Hough transform is introduecd as a universal method for object model matching using this multiple cell-features. This system can efficiently recognize objects unddr occlusion and noises. Model learning is performed by showing objects. In this paper, a system and hardware construction of vision system based on this engine is proposed. A prototype system demonstrates successful recognition of mechanical parts and equipment panels under uneven lighting-conditions and occlusion.