【摘要】 非下采样contourlet变换(nonsubsampled contourlet transform,NSCT)由于其平移不变性使得检测出的边缘定位准确。但是传统的基于非下采样contourlet变换的边缘检测结果中仍存在伪边缘,这是固定阈值选取不当所致。采用双阈值对高频子带中的模极大值进行筛选,用得到的两个矩阵进行补偿链接可以减少伪边缘。由于NSCT系数的结构特点,低频子带中也存在丰富的边缘信息,再用Canny算子对低频子带进行检测。仿真结果表明,基于双阈值的模极大值方法检测的高频子带细节丰富、定位准确、纹理清晰,Canny算子检测的低频图像边缘,轮廓完整连续,融合后的图像有效地抑制了噪声,消除了伪边缘。更多还原

【Abstract】 Due to the shift-invariance,nonsubsampled contourlet transform(NSCT) can accurately detect the edge orientation.However,traditional NSCT still reflects on pseudo-edge in the test result because of the improper selected fixed threshold.The dual-threshold is adopted to filter the modulus maximum in the high-frequency sub-band,and two matrices are obtained as link compensation in order to decrease the pseudo-edge.Owing to the structural characteristics of NSCT coefficient,there is also ample edge information in low-frequency sub-band,and then a Canny operator is used to detect the low-frequency sub-band.The simulation results show that there are abundant high-frequency sub-band details,accurated edge orientation,and clear texture,based on the method of modulus maxima of dual-threshold,and the Canny operator detects low-frequency image edge stability,figure integrity and continuation,and the fused image can control noise effectively and avoid pseudo-edge.更多还原