Frequency Domain Enhancement of Infrared Sensing Image Considering Multiscale Texture Features
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.
国家科技期刊平台
NSTL
万方数据
