关键词: 边缘检测, 各向异性高通滤波, 非下采样Contourlet 变换, 多尺度积

Abstract:

Image is always polluted by noise in processing. Since noise and edge are high frequency, noise is suppressed while edge is corrupted. In order to capture more accurate and true edges in noisy image, an edge detection algorithm is proposed by means of combining Anisotropic Highpass Filtering with multiscale product. Firstly, an original image is decomposed by Nonsubsampled Contourlet Transform, which is multiscale, multidirection and anisotropy. Secondly, the multiscale product is adopted in high frequency of transform in order to decrease noise. Finally, the relationship between anisotropic filter long axis and edge direction is introduced to locate image edges. The results of experiments show that the proposed method has strong denoising ability, low computational cost and be able to extract smooth and accurate edges. 

Key words: edge detection;anisotropic highpass filtering;NSCT;multiscale product