考虑多尺度纹理特征的红外传感图像频域增强  

Frequency Domain Enhancement of Infrared Sensing Image Considering Multiscale Texture Features

作　　者：曾琪[1] 杨真[2] ZENG Qi;YANG Zhen(Department of Information Engineering,Jiangxi Technical College of Manufacturing,Nanchang Jiangxi 330095,China;Network&Information Center,East China Jiaotong University,Nanchang Jiangxi 330013,China)

机构地区：[1]江西制造职业技术学院信息工程学院,江西南昌330095 [2]华东交通大学网络信息中心,江西南昌330013

出　　处：《传感技术学报》2024年第4期652-657,共6页Chinese Journal of Sensors and Actuators

基　　金：江西省教育厅科学技术研究项目(GJJ209928)。

摘　　要：红外传感图像质量容易受探测器和传输距离影响,导致图像亮度和对比度较低、轮廓细节模糊等问题。为此,提出了考虑多尺度纹理特征的红外传感图像频域增强方法。引入残差学习策略,基于多尺度纹理特征搭建多尺度卷积神经网络模型,进行图像去噪。对去噪后图像进行傅里叶变换,获取红外传感图像的低频图像和高频图像。针对低频图像部分,调节图像灰度和对比度以增强低频分量。针对高频图像部分,利用Log算子和Laplace算子增强图像细节及边缘。加权融合两者处理结果,选取Gamma校正调节对比度,增强高频分量。融合两种增强后图像,实现红外传感图像频域增强。实验结果表明,该方法峰值信噪比高于43,信息熵大于8,边缘强度超过82,对比度熵大于8.1,平均梯度大于8。The quality of infrared sensing image is easily affected by the detector and transmission distance,resulting in low image bright-ness and contrast,blurred contour details and other problems.Therefore,a frequency domain enhancement method of infrared sensing image considering multi-scale texture features is proposed.The residual learning strategy is introduced to build a multi-scale convolution neural network model based on multi-scale texture features for image denoising.The low frequency image and high frequency image of the infrared sensing image are obtained by Fourier transform of the denoised image.For the low-frequency image part,the image gray and con-trast are adjusted to enhance the low-frequency component.For high-frequency image,Log operator and Laplace operator are used to en-hance image details and edges.Weighted fusion of the two processing results is performed,and gamma correction is selected to adjust con-trast and enhance high-frequency components.The two enhanced images are fused to realize the frequency domain enhancement of infrared sensing image.The experimental results show that the PSNR of the proposed method is higher than 43,the information entropy is greater than 8,the edge intensity is greater than 82,the contrast entropy is greater than 8.1,and the average gradient is greater than 8.

关 键 词：多尺度纹理特征 红外传感图像 图像频域增强 卷积神经网络 GAMMA校正 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]