Retina-inspired neuromorphic edge enhancing and edge detection

摘要

In this paper, a novel analog retinomorphic block performing the edge enhancing and edge detection caused by lateral inhibition phenomenon is proposed. This phenomenon, which occurs in human retina, causes the visual acuity to improve in the edge regions of the object. In contrast to the negative meaning of the word "inhibition", this type of behavior causes the human eye to act as an analogue image processing chip. The reason for that is the edges of object are enhanced and the contrast ratios in the edge regions of an object are increased owing to the lateral inhibition. We adapt the process of convolution and the concept of image masking used in digital image signal processing method to analog image signal processing method. Our proposed circuit, which exhibits edge enhancing behavior thanks to lateral inhibition, consists of only current mirrors and current subtractor circuits. We obtain the analysis results via a simple circuit design having lateral inhibition feature and the results are quite similar to that occurring in human retina. In addition to the ability of lateral inhibition, we utilize from the masking property of the lateral inhibition circuit and change the coefficients of the mask in order to obtain an edge detecting circuit. For this purpose, we use a mask whose sum of the coefficients is equal to zero. Two different 500 x 500 pixel silicon networks are designed for both edge enhancement and edge detection circuits. We have analyzed an image with the helped of both edge enhancement and edge detection circuits. Analyses results reveal that our suggested block, which is set up using only MOS transistors, can enable edge enhancement and edge detection in grayscale image and lateral inhibition behavior of the human retina can be provided using a simple electronic circuit design which has the ability of performing convolution operation. TSMC CMOS 0.18 mu m process model is utilized to simulate proposed two analog circuit networks. (C) 2019 Elsevier GmbH. All rights reserved.