DESIGN OF A PIXEL-PARALLEL FEATURE EXTRACTION VLSI SYSTEM FOR BIOLOGICALLY-INSPIRED OBJECT RECOGNITION METHODS

This paper proposes a biologically-inspired feature extraction method, which consists of coarse region segmentation by a resistive-fuse network and feature extraction by Gabor wavelet transforms. Their pixel-parallel VLSI implementation based on the pulse modulation circuit architecture is described, and measurement results of Gabor filter operation by a test LSI chip with 1-D 20-pixels are presented.