Algorithm improves color edge detection

Image contours are often used by machine-vision systems to locate objects. In looking for contours in color images, most algorithms apply gradient information to each red (R), green (G), and blue (B) channel and then combine these contours to obtain an edge gradient. But this method is time-connsuming and complex.

Algorithm improves color edge detection

Image contours are often used by machine-vision systems to locate objects. In looking for contours in color images, most algorithms apply gradient information to each red (R), green (G), and blue (B) channel and then combine these contours to obtain an edge gradient. But this method is time-connsuming and complex.

A more elegant method that combines the information from R, G, and B channels before finding the contours has been developed by Aldo Cumani and colleagues at the Computer Vision Laboratory of the Instituto Elettrotechnico Nazionale (Turin, Italy).

The quadratic function that Cumani and his colleagues have developed generalizes the concept of edge detection in gray-level images to the multispectral case. After contour lines are found they are approximated with simpler curves, and the color attributes that describe the average image value on both sides of the contour approximated. These attributes are evaluated by analyzing the changes in the multispectral contrast value of the contours with different amounts of Gaussian blurring applied to the image. In this way, the problem of finding separate contour lines in each R, G, and B channel is avoided. For more information contact denasi@is.ien.it.

More in Boards & Software