Bandwidth reduction of multichannel images for machine vision

DETAILED DESCRIPTION The foregoing objects are among those met by the invention which provides machine vision methods for reducing the bandwidth of multichannel images, e.
g.
, based on contrasts in selected locations of a training image.
In one aspect of the invention, the method calls for inputting a multichannel training image, where each channel represents a respective (albeit, possibly overlapping) spectral band of the image.
The method identifies, e.
g.
, via operator input or otherwise, one or more calibration locations in the image.
The method determines weighting factors for each respective channel based on the contrast in that channel at a location corresponding to the calibration location.
The weighting factors are then used to reduce the bandwidth of a runtime image, which may be the same as or different from the training image.
Particularly, a reduced bandwidth runtime image signal is generated as a function of (i) the weighting factors determined from the training image and (ii) the multichannel runtime image signal.
By generating the weighting factors based on contrast at selected locations in the training image, the invention permits the preservation, enhancement, or reduction of selected contrasts in the reduced bandwidth runtime image signal.
To illustrate, by way of example, the invention can be used to facilitate bandwidth reduction of color images in the automated assembly of printed circuit boards.
Thus, a multichannel color image of a printed circuit board on a conveyor belt can be input as a training image.
A region of the image showing an edge of the circuit board can be selected, e.
g.
, by the operator, as a calibration location.
Weighting factors can be determined from the contrasts contained in each of the red, blue and green channels of the color image signal at regions corresponding to the calibration location.
A multichannel runtime image signal can be input, e.
g.
, showing several boards on a conveyor belt.
That signal can be converted to a single-channel, gray-scale image by multiplying each of the channels by its weighting factor and summing the results