结合无下采样Shearlet模极大值和改进尺度积的有噪图像边缘检测

Edge Detection of Noisy Images Combining Modulus Maxima with Improved Scale Product of Non-subsampled Shearlet Transform

为从噪声污染的图像中提取出更为清晰、连续的边缘,进一步改善边缘检测效果,本文提出了一种基于无下采样Shearlet模极大值和改进尺度积的边缘检测方法。首先对含噪图像进行多尺度、多方向无下采样Shearlet变换(Non-subsampled Shearlet Transform,NSST),得到图像在NSST域的高频系数;然后选取相邻的两个较大尺度的高频系数进行改进的尺度积运算,并经NSST模极大值处理得到边缘二值图像;最后使用区域连通方法去除二值图像中的孤立点,得到准确的边缘图像。大量实验结果表明,与小波模极大值、小波尺度积、基于无下采样Contourlet变换(Non-subsampled Contourlet Transform,NSCT)模极大值和尺度积、NSST模极大值等4种边缘检测方法相比,本文提出的方法具有更强的抗噪能力,且有效地避免了纹理的影响,检测出的边缘完整清晰,连续性好。

In order to extract clearer and more continuous edges from a noise-contaminated image,and further improve the effect of edge detection,a method of edge detection is proposed based on modulus maxima and improved scale product of non-subsampled shearlet transform.Firstly,the multi-scale and multi-direction decomposition of noise-contaminated im-age is performed through non-subsampled shearlet transform （NSST）to get the high-frequency coefficients in NSST do-main.Then the high-frequency coefficients in two adjacent larger scales are selected for the improved multi-scale product operations,and the NSST modulus maxima processing is carried out to get the binary image of edges.Finally,the isolated＆amp;nbsp;points are removed according to the regional connectivity and the accurate edge image is obtained.The simulative experi-mental results show that,compared with four edge detection methods （wavelet modulus maxima method,wavelet scale product method,non-subsampled contourlet transform （NSCT）modulus maxima and scale product method,NSST modulus maxima method）,the method proposed in this paper has stronger ability of resisting noise,and the influence of texture is a-voided effectively.The detected edges are clear,complete and with high continuity.