Boundary Detection Based on Chromatic Color Difference and Morphological Texture Suppression

A low computational cost technique is presented for boundary detection inspired by the human visual system perception. The current contributions are a chromatic color difference color model, a morphological texture suppression filtering, and a neighborhood related edge thinning. The proposed dRdGdB color model defines a RGB inter-component color difference, and showed a significant improvement in the performance of edge detection compared to RGB, CIELAB, YCbCr, HSV and YIQ. These improvements turn a prior edge detector into a boundary detector, allowing high confidence on separating external and internal objects´ edges. The segmented images perform well on the Berkeley Segmentation Dataset and Benchmark, achieving state-of-the-art performance.