Improved Contour Detection by Non-classical Receptive Field Inhibition

We propose a biologically motivated computational step, called nonclassical receptive field (non-CRF) inhibition, to improve the performance of contour detectors. We introduce a Gabor energy operator augmented with non-CRF inhibition, which we call the bar cell operator. We use natural images with associated ground truth edge maps to assess the performance of the proposed operator regarding the detection of object contours while suppressing texture edges. The bar cell operator consistently outperforms the Canny edge detector.