Research On Image Super-resolution Reconstruction Algorithm Based On Wavelet Domain

As a classic problem in image field,image resolution has attracted more and more attention in recent years.Image resolution determines the amount of information contained in an image.Therefore,improving image resolution is of great significance for information storage and utilization.Image super-resolution reconstruction mainly uses software method to improve the image resolution,and does not need to spend expensive money to complete the hardware improvement,which reduces the research cost and broadens the application field.It is widely used in medicine,remote sensing and public security.In the process of low-resolution image reconstruction,the traditional deep learning method does not distinguish the low-frequency structure information from the high-frequency detail information.There is a lack of information exchange between the layers of the network.The image features are only obtained through a single convolution kernel,and the image features are processed equally,so the feature utilization is seriously insufficient.This results in the loss of information in the results of high-resolution reconstruction image.In order to obtain more structure and detail information of image features,this paper combines wavelet transform with residuals dense network,which greatly increases the sparsity of the network,at the same time,fully excavates image feature information and improves the representation ability of the model.At the same time,by adding a multi-scale feature fusion module,the algorithm can not only obtain a variety of features in the same layer,but also increase the nonlinear expression ability of the network,and realize the deep mining of low resolution image information.In addition,channel attention mechanism is introduced in depth feature extraction.By adjusting the weight of feature graph between channels,channel correlation is increased,and more feature information with stronger representation ability is extracted.Finally,the sub-pixel convolution is used to avoid the damage of structure information caused by linear interpolation.In this paper,div2k data set is used as training data set,and set5,set 14,bsds100 and urban100 data sets are used as comparative test data set.At the same time,the algorithm is tested on the elm purple leaf beetle data set to verify the universality of the algorithm.The experimental results verify the effectiveness of the proposed image super-resolution reconstruction algorithm in the test set.It can make more efficient use of the original image information and solve the problem of missing information,so that the texture of the reconstructed image is clearer,the details are richer and the visual effect is better.