Edge Detection of High-Resolution Remote Sensing Image Based on Multi-Directional Improved Sobel Operator

Abstract

Automatic detection of building edge information from high-resolution remote sensing images can more accurately obtain building distribution information, which is of great significance for urban land planning, urban building planning and population estimation. The satellite image will be affected by atmospheric interference and noise, we preprocess the image to eliminate the atmospheric interference such as atmospheric correction and radiometric calibration, and utilize a new total variation and wavelet adaptive thresholding hybrid filter proposed in this paper to achieve the filtering of noise while retaining the image edge detail information. The traditional detection operator is prone to the problems of scattered edge points, discontinuity or misjudgment of too many edge points in the detection results due to the limitations of directions, template sizes. Therefore, this paper proposes a new 16-direction 5  5 size Sobel operator to replace the previous Sobel operator with only horizontal and vertical directions. These 16 directions are utilized to calculate the gradient of each pixel points. The templates for each direction are extracted for weighted fusion into the contour image for each direction. Thus, the edge information of different angles can be extracted, and the all-round information edge extraction of the image can be realized. By comparing the experimental visual effect and the evaluation criteria of the data results, the SNR, PSNR, AUC and FOM values of the improved algorithm are much higher than other algorithms, with the advantage of higher image detection edge positioning accuracy, more complete contour lines and stronger anti-interference ability.