Research On Wavelet Transform And Deep Learning Super-resolution Algorithm For Single Image

The super-resolution(SR)technology of images has been widely used in many fields such as traffic safety supervision,agricultural environmental supervision,medical and health care,remote sensing satellites,criminal detection and pattern recognition.SR technology is used to process Low Resolution(LR)images in order to reconstruct High Resolution(HR)images from LR image.By using computer software algorithm to restore the LR image to a richer and clearer HR image with texture details,effectively improving image quality and providing technical support for related fields,it is a more economical and feasible technical means.The SR technique of single image is one of the most basic and important contents in image super-resolution research.With the rapid development of deep learning in various fields of artificial intelligence and image processing,super-resolution technology based on deep learning has become a hot topic in current research.However,the SR method based on the combination of interpolation algorithm,wavelet transform and deep learning is not systematic and comprehensive.This paper studies the single image SR method through theoretical research,experimental analysis and application.The main work of this paper:firstly,the relevant theoretical basis of super-resolution algorithm is sorted out,the existing digital image super-resolution algorithm is classified,and the relevant key technologies are elaborated Secondly,previous studies have studied image quality reduction independently and separately from interpolation method,which means that interpolation algorithm cannot make full use of the quality reduction model.Different from previous research methods,this paper closely combines image quality reduction process with interpolation algorithm.First,it elaborates how to use different wavelet basis as the quality reduction kernel for quality reduction,and then discusses the interpolation effect relationship between different traditional interpolation algorithms and various quality reduction cores.The SR algorithm combined with wavelet transform and three kinds of interpolation can improve the image quality and achieve satisfactory results.Thirdly,with the help of Deep learning theory,using the current scholars put forward technology of Super Resolution algorithm of Deep network model(Very Deep Super Resolution,VDSR),an improved deep learning model is proposed,which combines three different interpolation methods and improved network models with different network layers.A super-resolution reconstruction algorithm for digital images is constructed.Fourthly,Combining wavelet transform with VDSR,the effect of interpolation and deep learning in different wavelet domain sub-bands on the effect of super-resolution algorithm is studied,and the related parameter optimization in the algorithm is discussed.Fifthly,The proposed super-resolution algorithm is applied to a traffic supervision and management system,which can improve the image quality.Through the platform,the image color is brighter and the edges are clearer.The related research provides a technical support for traffic safety.