Research On Multi-example Feature-constrained Back-projection Method Forimage-resolution

The purpose of image super-resolution is to get a clear high-resolution image from one or a group of low resolution images.As a basic step in the image processing,it is widely used in many fields,such as computer vision,medical imaging,computer simulation,and digital media technology.Because of the need to estimate high-resolution pixels from a group of pixels in the low-resolution images,image super-resolution still faces many challenges.The high frequency information such as edge and texture can generally accurately convey the effective information which described by the image.So how to maintain the high frequency information of the super-resolution image has always been a hot and difficult problem in the field of image super-resolution.The traditional image interpolation algorithm is simple,fast and easy to operate,and it has been applied in many fields of image processing.But traditional interpolation algorithms often cause fuzzy details and edge jagged effects.In recent years,image super-resolution algorithm based on example learning has also become popular.This kind of algorithm is to predict unknown pixel of high resolution images by learning from known instances.Compared with the traditional image interpolation algorithm,the algorithm based on example learning often has a higher complexity,but the resulting super-resolution image can keep better visual effects,retain more image features and more image details.Therefore,this paper combines the traditional interpolation algorithm with the example-based algorithm,and uses a simple and efficient interpolation algorithm to initialize the high-resolution image.At the same time,the similarity relationship between the pixel patches is fully considered,and the learnt model between the similar patches of the image is used to solve the problem of fuzzy details caused by interpolation,so that we can get high-resolution image with better visual effects.The classical example-based algorithm generally initializes high-resolution image by selecting a relatively simple polynomial interpolation algorithm,considering the convenience of calculation.However,as we mentioned above,the traditional polynomial interpolation algorithm always leads to the effect of blurred image and the jagged effect along the edge,since the feature area of the image is not fully considered.In this paper,a simple and efficient algorithm of feature constraint polynomial interpolation is proposed to initialize high resolution image.As the basis of the whole process of image processing,high quality initialization of high-resolution image has a direct impact on the final image.After that,we use the similarity of the image to get the learning model,and the learning model is used to find the high frequency information that initialization image lost.Finally,we use the iterative back-projection algorithm as a global post-processing means to get more accurate high-resolution image.During the experiment,medical images and natural images are selected as the experimental samples.Compared with the representative super-resolution algorithm in recent years,the results show that,the proposed algorithm can achieve better visual effects,especially in high frequency regions such as texture and edge.In terms of quantitative data,the algorithm proposed in this paper is also significantly improved compared with other algorithms.