A PWM-Mode Pixel-Parallel Image-Processing Circuit Performing Directional State-Propagation and Its Application to Subjective Contour Generation

摘要

As an approach to constructing highly organized and intelligent image recognition systems mimicking the visual architecture in the human brain, we have designed an image processing circuit that performs a directional pixel-state propagation algorithm based on a pixel-parallel architecture using pulse-width modulation (PWM) signals. Directional pixel-state propagation can be used for subjective contour generation, which is a typical function of the visual system in the human brain to fill gaps in natural-image information. A PWM-mode circuit can provide more compact circuit configuration than digital circuits, and can achieve focal-plane analog processing by combining with image sensors without A/D converters. Such circuits are suitable for repetitive algorithms, which require offset-free arithmetic. The proof-of-concept LSI chip has been designed and fabricated using a 0.25 m mixed-signal CMOS process. There are 35 35 processing units included in the chip, the operation cycle time for propagation step is 2.2 s, and the power consumption is 363 mW at a supply voltage of 3.3 V. In experiments using this LSI chip, we have successfully generated subjective contours of Kanizsa figures.