3-LEVEL BROAD-EDGE TEMPLATE MATCHING AND ITS APPLICATION TO REAL-TIMEVISION SYSTEM

This paper presents a new internal image representation, in which the scene is encoded into a three-intensity-level image. This representati on is generated by Laplacian-Gaussian filtering followed by dual-thres holding. We refer to this image as three-level broad-edge representati on. It supresses the high frequency noise and shading in the image and encodes the sign of relative intensity of a pixel compared with surro unding region. Image model search based on cross correlation using thi s representation is as reliable as the one based on gray normalized co rrelation, while it reduces the computational cost by 50 times. We exa mined the reliability and realtime performance of this method when it is applied to an industrial object recognition task. Our prototype sys tem achieves 32 x 32 image model search from the 128 x 128 pixel area in 2 milli-seconds with a 9 MHz pixel clock image processor. This spee d is fast enough for searching and tracking a single object at video f rame rate.