基于自适应多尺度积阈值的钢轨裂纹红外图像增强

Using Adaptive Multiscale Products Thresholding to Enhance the Infrared Image of a Rail Crack

利用红外热波技术检测钢轨斜裂纹伤损所得的红外图像存在整体亮度偏暗、边缘对比度低和纹理细节模糊的缺点,对测量裂纹的扩展深度不利。对此,本文提出一种基于非子采样Contourlet变换NSCT(Nonsubsampled Contourlet Transform)的自适应多尺度积阈值AMPT(Adaptive Multiscale Products Thresholding)钢轨裂纹红外图像增强算法。对图像进行NSCT分解得到不同尺度不同方向上的子带系数,由子带系数自适应地确定阈值并调整增强函数,将阈值作用于每一子带的多尺度积,对阈值化操作后的子带系数进行增强处理和逆变换,实现图像增强。实验结果表明:该算法在抑制噪声和提高图像整体对比度的基础上,有效突出了目标图像的边缘,对裂纹内部微弱细节纹理的增强效果优于其他传统的红外图像增强方法,为进一步提取代表裂纹深度的关键像素提供了有力支撑。

The infrared image obtained from the use of infrared thermal wave technology to detect an oblique crack of a rail has a number of shortcomings,including a low overall brightness,low edge contrast and blurred minutiae texture,which would result in detrimental impact on the measurement of the depth of the crack expansion.Therefore in this paper,a nonsubsampled Contourlet transform（NSCT）based adaptive multiscale products thresholding（AMPT）algorithm is proposed to enhance the infrared image of a rail crack.Firstly,the image was decomposed using the NSCT to obtain multi-scale and multi-direction sub-band coefficients,which then adaptively selected the threshold values and adjusted the enhancement function;subsequently the threshold was applied to the multi-scale product of each sub-band;and finally the thresholding processed sub-band coefficients were enhanced and the results underwent a further inverse transformation,thereby realizing the image enhancement.The results of the experiment demonstrate that the proposed algorithm can suppress the noise,increase the overall image contrast,while effectively highlight the edge of the target image.This approach is superior in inner-crack minutiae texture enhancement to traditional infrared image enhancing methods,therefore providing apowerful support for the further extraction of key pixels that can characterize the depth of a crack.