Hybrid Implementation of Marr-Hildreth Method of Edge Detection

Abstract

A new hybrid method for object edge detection is suggested which uses a combination of analog and digital processing. Detection is performed using second derivative approach known as Marr-Hildreth method. Smoothing and physical measurement of gradient is performed at the imaging step in hardware of linear array of differential sensing cells. It is followed by digital computation of divergence of gradient. The result of imaging of sheet-like media is differential image, where one of components of gradient is recorded directly while for the other a single-band modulation of spatial carrier frequency is used. Point spread function (PSF) of the imaging array cell consists of negative and positive lobes of triangular shape oriented in opposite directions. Such lobe arrangement provides direct analog measurement of gradient vector and spatial modulation of one of its components. The divergence of gradient is further calculated by digital filtering which implements demodulation of spatial carrier frequency, calculation of derivatives of gradient components, and their summation. Subpixel interpolation is finally performed to obtain Laplacian image of media. For the experiment an array of flat capacitive differential cells was used to detect edges of objects formed as pieces of plasic film pasted to paper sheet. The size of PSF lobes of the cell was equal to 12 mm. In the Laplacian image acquired, the contour of increased thickness area of the sheet was marked from inside by a bright rim. Objects a few centimeters large were reproduced with correct size and slight distortion of the contour. The very presence of areas from 5 to 10 mm large was detected consistently, but their shape was heavily distorted and the size was bigger than in reality.