| The research of super-resolution technology is a hot topic in the field of image processing.It meets the modern technology’s intelligence,visualization and perceptible demand for machine equipment and information.Its research is powerful connection image analysis theory and artificial intelligence.In the future,with the further development of the research and the development of technology,the image superresolution technology faces a broader prospect..In this paper,an algorithm based on spatial information is proposed by combining three algorithms of PM model \Rayleigh scattering model,prediction residual Pyramid and structural tensor.The first innovation of this paper is to introduce the actual thermal diffusion model and the optical Rayleigh scattering model.Secondly,adaptive variable factors are introduced in the two models,so that the pixel in the model can be adjusted in a high frequency or low frequency.The key of the algorithm is how to set the parameters in the models so that it can be changed adaptively in magnification in the thermal diffusion model.In this paper,a meaningful attempt is to compute the mean and standard deviations of the divergence of adjacent pixels,and construct the coefficient K.This algorithm is more smooth than the linear amplification to maintain a smooth boundary.The third point is the introduction of pyramid algorithm to construct the reconstruction framework.This is a multi-scale analysis and multi-scale reconstruction method.Each layer of the pyramid is an extension of the previous layer of results.Each extension only has one pixel point to expand a 3X3 image block,avoiding the error caused by a large scale.The fourth point is that the super-resolution reconstruction algorithm based on the thermal diffusion model and the optical Rayleigh scattering model has the advantages of fast speed.There is no training and parameter adjustment compared to the neural network.In order to give full play to the advantages of convolutional neural network,this paper presents a new method to solve the limitation of low resolution optical image by combining with the first two methods.The convolution neural network here is essentially used for image enhancement to further optimize the above two algorithms.In this paper,the related algorithms of super-resolution image reconstruction are studied,and various algorithms and their applications are discussed in detail.The adaptive variable factor is applied to the thermal diffusion model and the optical Rayleigh scattering model for image reconstruction.An improved algorithm is also proposed,and the method is compared and analyzed by corresponding simulation experiments,which proves the superiority of the method. |
