A robust tensor-based edge detection method for noisy remote sensing images

Traditional gradient based edge detection methods such as Sobel, Laplacian of Gaussian and Canny detectors which have charming applications in common images are not satisfactory in dealing with noisy remote sensing images. Aiming to solve this problem, we introduce the trace of the nonlinear local structure tensor to sort edges out of noises. Then we extracted the contours of the trace and fill the gaps in the contours. At last, edge thinning is applied to the contour map to suppress the non-maximum pixels. Experiment results demonstrate that our robust tensor-based edge detection algorithm is effective under noisy condition.