Corner Detection Based On Image Edge Contour

Corner detection has always been an indispensable preprocessing process in digital image processing and computer vision.Some traditional detection algorithms use the curvature formula to calculate the curvature value of all pixels on the curve and take the local maximum curvature point as the candidate corner point.However,this kind of algorithm is very sensitive to quantization noise and detail change on the edge contour,which easily causes the instability of the detection result,thus affecting the performance of the corner detector.In view of the existing problems of this kind of detector,the paper studies the image corner detection algorithm,proposes a multi-scale point-to-chord distance corner detection algorithm based on image edge contour,and evaluates the performance through different experiments to verify the effectiveness of the algorithm.(1)Corner detection algorithm based on multi-scale point-to-chord distance.First,Canny edge detector was used to extract the edge map from the original image,and then each edge line was extracted.First,using Canny edge detector to extract the edge map from the original image,then extract the each edge contour,and using three different kinds of Gaussian scale to smooth edge contour,and connect the two end points of the curve to form a cord and computes the maximum Euclidean distance between the cord and the points of the curve,if it is larger than threshold and can be detected under three Gaussian scales,the pixel of the curves will be regarded as the candidate corner.Then connecting the corner to two ends of curve and continue to find the maximum point-to-chord distance,then all the obtained corner points whose maximum distance is higher than the threshold are marked as candidate corner points,until all candidate corner points are selected.Finally,applying the non-maximum suppression to candidate corner set and obtaining the final corners.Compared with the existing corner detection algorithm based on curvature calculation,the proposed algorithm does not need to calculate the first and second derivatives,and effectively avoids the calculation error caused by local changes.(2)The parameters involved in the algorithm are tested and analyzed,and the optimal parameter combination is obtained according to the performance evaluation criteria.16 standard test images were used to compare the average repeatability and localization error of Harris algorithm,He&Yung algorithm,CPDA algorithm and the algorithm proposed in this paper under rotation,uniform scaling,non-uniform scaling,shear transformation,JPEG quality compression and gaussian noise transformation.Experimental results show that the proposed algorithm has good performance under different image transformations.Two commonly images,'Block' and 'Lab',were used to test their noise robustness and corner matching experiments.According to the calculation,the average error detection rate of different detection algorithms in the case of no noise and noise is from high to low: CSS detection algorithm is 32.775%,Harris detection algorithm is 28.375%,He&Yung detection algorithm is 24.925%,CPDA detection algorithm is 22.925%,and the algorithm in this paper is 20.525%.Compared with other classical corner detection algorithms,the proposed algorithm has the lowest average error detection rate and better corner detection performance.