Research And Realization On Classification Of Image Degraded Factor

The process of obtaining the imaging data of space images will be affected by many factors.For example,the cameras,imaging platforms,imaging environment,andthe characteristics of the targets themselves are all the key factors influencing the images' quality.Targets and the dynamic characteristics of the cameras,imaging platforms' working modes,the imaging environment and the imaging results will produce certain effect.The sequence of images and the information contained in images are virous under different influence factors,which will enhance the difficulties in images' processing.Therefore,detailed analysis of features of each drop qualitative factors and the existing form can provide theoretical support for image quality evaluation and designing the parametric modeling of reduction factor methods.In this paper,the purpose is using support vector machine(SVM)and the typical mode of deep learning,convolution neural network to study classification method of images with qualitative factors.First of all,analyze the reasons for the formation and forms of existence of the drop qualitative factors,and their influence during the imaging process of the image datas,which can be taken as principles to classify datas and bulid target database,then build composite reduction image database and single mass reduction image database;Then select airspace and frequency domain characteristics of images,mainly includes wavelet transform,gradient characteristics,use these features constitute characteristics vector to sense the form of qualitative factors ofa image,,and use it as the input of the support vector machine(SVM)and deep learning of the neural network,through the study of the settings of the parameters of the support vector machine(SVM),then use support vector machine(SVM)to realize coarse classification of composite factors for images;choose deep learning network model,Alex.net,analyzes its network principle,algorithm implementation details,each layer of the data transmission mechanism,the network training process detailedly,use its strong self-learning ability,to realize the kind of a single factor by learning.The main research and innovations are as follows:1)Analyzed the reasons for the formation and the form of existence of degraded factors.Degraded factors play an important role when execute image quality evaluation and make image processing intelligent.Then built two image database.From the images' imaging link,this paper analyzed the reasons for the formation and forms of existence of the drop qualitative factors during the process of formation and types,and take them as principles to bulid image datas.For these cases that less chance imaging conditions,limited data,imaging characteristics,for different qualitative factors,by different ways such as download data,data simulation to establish qualitative factors image database and a single composite reduction factor of image database,which offer this article's classification of degradation element and the image quality evaluation work to provide data basis in the future data support.2)Put forward a method to classify compound degradation element based on support vector machine(SVM).Based on the types of degradation element,the features of gradient and wavelet transform is adopted to form a characteristic vector of degradation element,perceptual image noise,fuzzy,high,low exposure mass reduction in image contrast,such as structure,details and so on.Then introduced the current commonly used methods of the classification and explain why choose the support vector machine(SVM).The effectiveness from experiments and statistical validation has certificated the method is reasonable,the acquire of all kinds of the factors' probability of qualitative provide a good basis for the space images' quality assessment and post-processing.3)Put forward a method to classify single degradation element based on the deep learning.Aiming at the case that there will be single degradation element in images,using the typical model of convolutional neural network(CNN),Alex net,who is good at realzing,take images that have single degradation element as the input of the network.After the original image going into the convolution layer,it can produce a characteristic figure,after entering the pooling layer for sampling,so on,finally the net will output a characteristics graph to form a column vector,the input to the classifier,the classifier can generally choose MLP(Multi-layer Perceptron Neural Networks)or Softmax classifier to output the final decision results,finally realizes the accurate classification of each drop qualitative factors.