Method Of Fog Automatic Identification Based On SVM Study

In recent years, the fog weather appeared in most areas of China and polluted our environment seriously, which has attracted more and more attention.The harms of fog mainly divided into two categories:one is the damage that produced to the human body,the other is the damage of transportation, especially on the road transport system that affected most seriously, even caused accidents directly or indirectly and resulted in threating to lives and property. Given the fog figure's impacts on human beings, the paper devoted to research on fog view of an automatic detection algorithm specifically,hope to set a precondition to improve the image quality of defogging, which can make its own contribution for the development and application of the image defogging methods. In this paper, two methods are put forward to detect whether there is fog in the image.(1)The fog figure detection method based on gradient texture descriptor: study from based on gray-gradient co-occurrence matrix method for discriminating an image fog,puts forward a new fog image determination method which is compatible with two characteristics of a large gradient strengths and gradient texture descriptors variance of the image,that can be determines whether there is fog in the image effectively.Through the acquisition of a large number of natural and non-fog fog picture image and combined with the artificial fog image texture features analysis, found that the texture features of the fog figure. Through analyzing and comparing a total of thirteen texture feature vectors, only four texture feature vector that affect the fog and fog-free images,respectively : contrast, angular second moment, high gradient strengths, gradient variance; then it put forward a new texture feature vectors :gradient haze, the feature vector solves the problem of fog figure that effected by the gradient well. Finally, a custom formula is used to detect the fog figure automatically.The experiments show that the method of fog figure detection based on gradient texture descriptors that proposed in this paper has a relatively high recognition rate.(2)The fog figure detection method based on support vector machine SVM :(support vector machine): The paper introduced the statistical learning theory firstly, which is a theory of machine learning regular that specifically designed for small samples. Then introduced two kinds of support vector machine : linear support vector machine and non-linear support vector machine. Finally, the gray level co-occurrence matrix method is used to extract image texture feature, that just need to divided the image into two categories: fog and without fog,so it is a relatively simple binary classification problem.The experiments show that the method of fog figure detection has a relatively high recognition rate...