Decision tree for corner detection

Corner detection is a important task in low level vision. Detecting corners helps one to establish similarity between two or more images. Traditional approaches for corner detection involve finding significant variation around a pixel neighbourhood in two different directions. In this work, we have developed a novel framework to detect corners in a given image by learning corners from images corresponding to the same object category. We detect extrema of the intensity and second derivative neighbourhood around a given pixel location to identify possible corners. We build a decision tree using the learned parameters and also employ the intensity variation in the local neighbourhood in order to detect corners accurately. We show that the performance of the proposed approach compares well with the standard corner detection algorithms and the other learning based approach for corner detection.