DESIGN OF A NEUROMORPHIC EDGE DETECTOR VISION CHIP WITH COLOR AND
INTENSITY CHANGE DISAMBIGUATION
ABSTRACT
A neuromorphic edge detector with color and intensity
change disambiguation using only one photo detector per
pixel has been developed. The proposed structure is very
simple, compact, and low power. It dissipates about
100nW per pixel with 1V power supply. It can handle a
wide range of input light luminosity and process image
streams up to about 30 frames per second. The detector
has been designed and simulated in TSMC 0.18um
CMOS process.
Index Terms—mage edge detection, Image color
analysis, Real-time image processing, Analog processing
circuits, Integrated circuits
1. INTRODUCTION
Real-time Edge detection is a fundamental tool in
image processing and machine vision. It is used as a
preprocessing step in many image understanding
applications. The traditionally frame-based approach to
deal with vision tasks is based on two following main
steps: 1) acquisition of the image data by a low-cost CCD
camera and 2) software processing of the data on a digital
platform (DSP or PC) [1]. The frame-based approach
carries some disadvantages due to its relatively high
power dissipation and high processor cost. An attractive
solution to this problem is to shift part of the image data
processing into the sensor hardware structure. It is done
by designing and allocating some circuitry around photo-
captor to perform some early pixel-level image
processing in the photo-sensor array [1]. In this way, the
CCD camera (or CMOS imager) is replaced with a more
sophisticated device which is able to preprocess the
acquired image data, and hence to provide the next
processing stages with less requirements in terms of
computational power, cost, portability and power
consummation [2]. One area of research is the use of
neuromorphic engineering techniques to implement
processing circuits inspired by biological neural systems.
In the present work, in-pixel processing circuitry of a
real-time neuromorphic edge detector with color and
density change disambiguation is introduced. The
proposed detector has been designed on the basis of retina
models in color and edge detection. Buried Triple
Junction photo detector (BTJ) is used as the pixel light
sensing element.
In the following, in section 2 retina structure is
explained. in Section 3, the proposed edge detector
system and its general basic blocks are explained. Section
4 describes in more detail the color and intensity change
detection blocks. Section 5 presents the simulation
results, and conclusion is provided in Section 6.
2. BIOLOGICAL INPIRATION
The retina is the first neural structure to be used in
biological visual data processing. It filters input signals
and extracts only the relevant information necessary in
the visual cortex for high level visual processing tasks[3].
Fig. 1 shows schematic cross section of the Macaque
retina containing the basic neural cells. The
photoreceptors convert incident light into electrical
signals. Each horizontal cell (H) spreads information over
neighboring horizontal cells and thus performs
spatiotemporal low-pass filtering on the photoreceptor
outputs. Bipolar cells (B) deliver the difference between
the receptor and horizontal cell responses. The difference
signal then will be encoded and transmitted through the
optical nerve and edges will be extracted from this
information. Trichromatic color vision is the ability of
humans and some other animals to see different colors by
possessing three independent channels for conveying
color information, derived from the three different cone
types. The three types of cones are L, M, and S, which
have pigments that respond best to light of long
(especially 560 nm), medium (530 nm), and short (420
nm) wavelengths respectively[4][5].
Fig. 1: Schematic cross section of the Macaque retina [3]
Hesam Ahmadi*, Sayed Masoud Sayedi**
*ECE, Isfahan University Of Technology, Isfahan 84156-83111, Iran,
h.ahmadi@ec.iut.ac.ir
** ECE, Isfahan University Of Technology, Isfahan 84156-83111, Iran,
m_sayedi@cc.iut.ac.ir
2012 25th IEEE Canadian Conference on Electrical and Computer Engineering (CCECE)
978-1-4673-1433-6/12/$31.00 ©2012 IEEE