On Selection and Combination of Relevant Color Components for Edge Detection

This paper aims to define the best suited combination of color components for edge detection. This is done by a selection of the most relevant components from each color space. First, we compare the performances of the conventional canny edge detector applied to this color spaces. Then, we tend to assess the accuracy of detected edges by using two quantitative methods namely, Fram and Deutsch in addition to robust statistical error measures. Experiments have been conducted on images from BSDS300 database using perceptual and correlated color spaces. We additionally perform these experiments on a progressively contaminated versions of each processed image. Relevant color components are defined according to their detection accuracy and gaussian noise robustness. The obtained results show that Y (luma of YIQ), Y (luminance of YCbCr) and V (brightness of HSV) perceptual color components remains the best candidates for an eventual combination.