Abstract
In the perspective of the spatial frequency domain, human visual system could superposition neighborhood pixel signal and enhance graphic edge, which is very similar to pixels through filter. Artificial retina chips need to provide high dynamic range and well edge detection for the blind persons to restore vision. CMOS image sensor technology has poor dynamic ranges than human eye. Fortunately, image processing would not only remedy this defect of dynamic range, but also strengthen recognitions of the graphic contour. Laplacian algorithm performs neighborhood processing with 3×3 weight mask. Using its characteristic of high sensitive to signals' difference to remove background light flux, thus effectively improve noise immunity for changes in light intensity. This thesis develops a hardware to implement Laplacian algorithm, and discusses its applications on artificial retina chips. Convolution algorithm based on a 3×3 kernel is realized by operational amplifier circuit with arithmetic weight function. The prototype circuit performs images enhanced by Laplacian algorithms in real time. The concept chip contains 64×64 pixel array and is implemented in TSMC 0.18um Mixed Signal Technology. Every array element mainly consists of pixel unit embedded with Laplacian algorithm element.