Research And Implementation Of Image Super.resolution Reconstruction Based On Deep Learning

As one of the most common information carriers in our daily life,images are affected by external factors in the process of generation,which makes the final image quality often unable to meet people's requirements,and more or less has an impact on people's lives.Therefore,how to improve image quality and clarity has become an important research direction in the field of image processing.The technology of image super.resolution has been successfully applied to computer vision,medical image processing,satellite image,video surveillance and so on.The aim of image super.resolution reconstruction is to improve image quality.The methods of image super.resolution can be divided into hardware and software.It is too expensive to generate high.resolution images by improving hardware technology,so the practical and effective method is to acquire high.resolution images by means of software algorithm.In recent years,with the rapid development of artificial intelligence,all walks of life have entered the era of artificial intelligence.Machine learning and neural network algorithms are also widely used in the field of image processing.Therefore,how to transform low.resolution image into high.resolution image by machine learning and neural network has become an important subject in the field of image processing.The main works of this paper are as follows:(1)After experimenting on the image super.resolution algorithm based on convolution neural network,this paper first improves the method of feature extraction in image reconstruction process to reduce the loss of information in feature extraction.Using three.layer convolution to process image,and introducing pooling layer to reduce the output eigenvectors of convolution layer,reduce dimensions,improve training efficiency,deepen the network level helps to learn more essential expression of image data.Through a series of comparative experiments,the effect of the improved algorithm is verified,and the network performance is analyzed according to the network parameters.Finally,two important objective evaluation criteria,peak signal.to.noise ratio and structural similarity,are used to evaluate the reconstructed image quality.The experimental results show that the image reconstructed by the improved algorithm achieves good results both in subjective perception and objective evaluation criteria.(2)For image super.resolution reconstruction system,a system of image super.resolution reconstruction on WEB side is proposed and implemented.On windows platform,the front page of the system is designed and built by using Django development environment and python GUI(Tkinter)graphical user interface,and the background database of the system is designed and implemented by using SQL server database.In this paper,the system is applied to video surveillance to facilitate staff to quickly find suspicious people and vehicles or other items through surveillance.