Texture-suppressing edge detection in real-time

The problem with real-time edge detectors today is their sensitivity to the unimportant edges in textured regions. Even the industry standard Canny edge detector suffers from this problem. At the same time, texture boundary detectors cannot run in real time as they are too computationally intensive. We propose an edge detector that is able to suppress the unimportant edges found within a texture, while being able to locate the important edges within an image in real time. Tests show it is able to run consistently at 20.75 frames per second, making it fast enough to be considered real-time, while impressively scoring an F-statistic of 0.59 on the Berkeley segmentation benchmark.