Research On Adaptive Image Fusion Algorithm Based On Quality-driven

Image fusion is an important branch in the field of image processing.It can process image data comed from different types of sensors which collect information of the same scene,it retains useful information from different images and removes redundant information,and gets high-quality images containing integrated information.Image fusion plays an important role in both military and civilian fields.Deep learning is developing rapidly and has a very wide range of applications in the field of image processing,but in the field of image fusion,deep learning methods do not have obvious advantages over traditional methods.There are two main reasons: first,there is no optimal fusion result in the field of image fusion,resulting no available ground truth for deep learning,making it difficult to make full use of the advantages of the strong fitting ability of the network.Second,it is very important to judge salient regions,and to deal with high-frequency information and low-frequency information in different ways in image fusion,which are difficult to achieve in an end-to-end deep learning model,which is difficult to reproduce in an end-to-end deep learning model.But the deep learning method still has its' unique advantages.Deep learning can automatically extract the feature information inside the image without setting the rules of feature processing;deep learning has strong system adaptability,and the model trained can be improved and optimized easily for the new task.Therefore,this thesis explores and studies the image fusion method based on deep learning,The main research contents are as follows:First of all,this paper investigates the related research works of image fusion,introduces several sensor images commonly used in the field of image fusion,sorts out traditional pixel-level image fusion methods,analyzes the advantages and disadvantages of image fusion methods based on deep learning,and introduces the commonly used image fusion quality evaluation indicators.Then,with the fundational work of research on image fusion method which is based on deep learning,an image fusion method based on multi-layer generative adversarial networks is proposed.Inspired by Fusion GAN,this mothod decomposes the input fusion images into a multi-scale images and uses the multi-kayer network to fuse them,so that the network can fully distinguish and learn the information from global scene and details,making the final fusion image results have more details and clearer edges than Fusion GAN.Compared with Fusion GAN and a variety of traditional image fusion methods,the superiority of the mothod proposed in this paper is verified.Finally,in this thesis,a quality-driven image fusion method based on the neural network model is proposed.By constructing loss functions on image fusion quality evaluation indicators,the parameters of the image fusion network are continuously optimized to solve the difference of input images from the training data set,resulting in excellent fusion quality of the output images.This is a adaptive image fusion method based on quality-driven.