Research On Super-resolution Reconstruction And Spatial Fusion Enhancement Algorithms For Uneven Light Images

Affected by climate,environment,and imaging equipment,the acquired images often have problems such as uneven illumination,excessive noise and blurred edges,which lead to the inability to effectively extract images' features and affect the expression of important features.Therefore,this paper selects the downhole image as the research object of uneven illumination,and improves a super-resolution reconstruction algorithm of uneven illumination image.To solve the local fuzzy problem of reconstructed images,a space fusion enhancement algorithm is improved.The main work of this paper is as follows:(1)In order to solve the problem of image distortion and edge blurring in the traditional super-resolution reconstruction algorithm,an improved super-resolution reconstruction method is proposed based on the original extremely deep super-resolution(VDSR)reconstruction method.First of all,a self-gated Swish activation function is used to replace ReLU activation function,which effectively solves the overfitting problem caused by too many network layers and prevents the occurrence of invalid convolution.Then,an improved residual network is used in the network structure,which can not only ensure the number of network.layers,but also learn more in-depth image details,which solves the problem of incomplete and missing information feature extraction due to multiple image transmission in VDSR,and can better learn end-to-end mapping.Finally,high-resolution images are obtained using deconvolution at the end of the network.Experiments show that the improved reconstruction model can obtain higher objective evaluation index.(2)A space fusion enhancement algorithm is improved to solve the problem of local edge blurriness after illumination uneven image reconstruction.The reconstructed image is converted from grayscale image to RGB image,then RGB is converted to HSV image and brightness component is extracted.The luminance component is denoised by guided filtering and the image edge feature is extracted.The extracted feature image and the processed luminance component are decomposed by non-sub sampled shear wave(NSST).The high-frequency subbands are fused using the PCNN model,and the high-frequency features are used as the model excitation.The total ignition amplitude is maximized to fuse the high-frequency features of the image;the low-frequency subbands are fused using an improved energy maximization method.The edge information has been effectively protected and integrated.Finally,the inverse transform of NSST is performed,and the inverse transform of HSV is performed with the H and S components to obtain an enhanced image.Experiments show that the improved fusion algorithm enhances the image details and achieves good results in objective evaluation.