SPIE Digital Library Get updates from SPIE Newsroom
  • Newsroom Home
  • Astronomy
  • Biomedical Optics & Medical Imaging
  • Defense & Security
  • Electronic Imaging & Signal Processing
  • Illumination & Displays
  • Lasers & Sources
  • Micro/Nano Lithography
  • Nanotechnology
  • Optical Design & Engineering
  • Optoelectronics & Communications
  • Remote Sensing
  • Sensing & Measurement
  • Solar & Alternative Energy
  • Sign up for Newsroom E-Alerts
  • Information for:
SPIE Defense + Commercial Sensing 2017 | Register Today

OPIE 2017

OPIC 2017




Print PageEmail Page

Electronic Imaging & Signal Processing

Machine vision

Optical correlation allows silicon retinas to recognize patterns stored in memory.

From oemagazine July 2001
30 July 2001, SPIE Newsroom. DOI: 10.1117/2.5200107.0008

Most machine vision systems separate the process of image acquisition from image analysis. Now, a programmable silicon retina can recognize patterns and evaluate the degree of resemblance of samples under test to stored patterns. The retina is based on the optical correlation between an image projected on the silicon substrate and a reference image memorized by an electronic device integrated on the same substrate.1-3 Built using standard 0.6-µm CMOS technology with three levels of metal, the circuit consists of a 100 X 100 pixel CMOS image sensor with an approximate size of 5.6 X 5.8 mm. The fill factor is greater than 30% for a pixel size of about 50 X 50 µm.

During a learning phase, the current produced by each of the illuminated pixels is compared with a fixed threshold value (Ith). Depending on whether the current is greater or smaller than Ith, the corresponding pixel is classified in the bright or dark region of the image. This operation partitions the array into two complementary disjoint subsets, yielding a binary correlation mask in which all the photodiodes in the array are connected to either one of the two outputs according to its classification into one of the two regions previously determined.

During the analysis phase, the retina furnishes two types of information in the form of electrical currents: a current proportional to the luminous flux falling on all the photodiodes pertaining to the white part of the mask (Iwhite), and a current due to the black part (Iblack). The projected image can thus be optically correlated to the reference image. When an image is observed, the signals at the output of the circuit constitute the set of parameters necessary to the classification process. They are compared with reference values, and the result of the comparison is a binary decision on the degree of resemblance of the two images. Our 100 * 100 pixel retina has been evaluated using a set of test patterns, and the results show that it is capable of detecting differences in the programmed and analyzed test patterns, or detecting spatial displacements in two identical patterns.

In most devices based on artificial vision, image analysis starts with a systematic scanning of all the pixels. With our retina, this operation is not necessary since the measure of the correlation value between the two images uses a purely optical process. This approach simplifies the integration of the device by requiring only passive components in conjunction with common logic circuits necessary for the classification and for the binary decision-taking, plus a monostable circuit that generates the command pulse for the memorization of the reference image. The optical process also minimizes response time, which allows the device to be used for solving pattern recognition problems in an environment with strong real-time constraints. Finally, it can be envisaged to extend its application to video surveillance of a premises by adding to it an external electronic system that allows the learning phase to be done at a regular rate. oe


1. P. Gorria, B. Lamalle, and G. Cathebras, "Dispositif d'intercorrélation d'une image," Patent No.99 00008, (1999).

2. B. Lamalle, P. Gorria, et al., Proc. EOS/SPIE Symposium on Applied Photonics, Glasgow, Scotland, UK, Vol. 4076, pp. 225-234 (2000).

3. L. Lew Yan Voon, G. Cathebras, et al., Proc. SPIE 4306 (2001).

L.F.C. Lew Yan Voon, B. Lamalle, P. Gorria, G. Cathebras, B. Bellach, F. Meriaudeau, and D. Navarro

L.F.C. Lew Yan Voon, B. Lamalle, P. Gorria, B. Bellach, F. Meriaudeau are with the Université de Bourgogne. G. Cathebras and D. Navarro are with the Université Montpellier II.