The Research On Image Super-resolution Restoration Algorithm Based On Gcforest

With high resolution requirements of video monitoring,medical,remote sensing and other fields on digital images,the super-resolution restoration technology of digital images has become a key research objective of the field of image processing.Because the digital image will be located in the way of acquisition,affected by all aspects of image degradation.In order to accommodate the needs of life and the requirements of scientific research,it is necessary to develop a super-resolution restoration technology to solve the problem of image degradation.It can recover a high-resolution image of one or more low-resolution images,and resolve the problems of practicality,complexity,low quality and long time of the current restoration algorithm.It can meet the high resolution requirements of people's living and scientific research fields and improve the performance and recovery efficiency of traditional algorithms.In this paper,an image super-resolution restoration algorithm based on deep forest is proposed.In this paper,the single-frame image super-resolution restoration is realized by improving cascade forest model and adopting multi-granularity scanning algorithm for feature extraction,so as to complete the structural optimization of the training model is enhanced the efficiency and performance of image restoration.To solve the problems of single frame image super-resolution restoration algorithm based on deep learning,such as less feature extraction,long restoration time and large amount of calculation,etc.Based on the divide-and-conquer strategy,a multi-frame super-resolution image restoration algorithm based on deep forest is proposed.The algorithm of multi-frame image registration is improved the decision tree.Accurate restoration of multi-frame images.The information contained in multi-frame sequence image is fully extracted to improve the quality of multi-frame image restoration.Solve the traditional multi-frame restoration quality is not high,complex structure and other problems.According to the experimental verification,the single-frame and multi-frame imaged restoration algorithm proposed for this paper has a great improvement in practicality and value compared with the current popular algorithm when the displacement of high-dimensional imaged data and multi-frame sequence images is large.The performance of traditional super-resolution image restoration algorithm is improved.The idea of super-resolution image restoration algorithm is extended.To meet the needs of life application and scientific research;promote thedevelopment of image processing.Pattern recognition and other disciplines.The main work is as follows:(1)enhance the information content of image features: this paper extracts features through multi-granularity scanning algorithm and multi-window scanning.Without repeated sampling,image feature vectors abounding in information can be obtained.(2)cascade forest learning model improves the quality of the restored images: the cascade forest model composed of complete random forest and ordinary random forest is used for multi-level circular training.High resolution image blocks recovered from each layer are taken as the input of the next layer,and the high resolution image blocks recovered from the front and rear layers are compared to determine whether the training is finished.(3)super-resolution image restoration algorithm based on the depth of the forest is: adopting multi-granularity scanning algorithm for input of low resolution and high resolution image block to feature extraction,respectively,in the cascade of cascade forest model way iterative training characteristics,through the leaf node of high and low resolution image block the mapping relationship to get regression model.The regression model was used only for image restoration.(4)registration algorithm based on decision tree: by improving the algorithm of feature extraction in the registration process,this paper solves the problem of less feature extraction in traditional registration and combines EM algorithm for accurate registration.Improve the accuracy of low resolution image registration.The practicability and value of multi-frame super-resolution restoration algorithm based on depth forest model are enhanced.