一种基于SVM分类的雾图自动检测方法  

An Automatically Detection Method of Hazing Images Based on SVM Classification

作　　者：胡众义[1] 刘清[1,2] 郭建明[2] 徐华中[2] 

机构地区：[1]武汉理工大学智能交通系统研究中心,湖北武汉430063 [2]武汉理工大学自动化学院,湖北武汉430070

出　　处：《计算机仿真》2015年第2期342-346,共5页Computer Simulation

基　　金：国家自然基金项目(51279152);浙江省自然基金项目(LY12F02015)

摘　　要：从图像中自动检测出是否有雾是智能去雾的前提。为此提出一种从室外自然图像中检测有雾图像的方法,把图像的RGB分量和亮度进行规范化处理,使得图像与外界光线强度关联最小。然后通过二维离散傅立叶变换获得图像频谱特征,再结合图像的灰度共生矩一起作为分类的特征。收集大量室外彩色无雾图像,进行模拟加雾,通过训练和评估,获取检测有雾图像的矢量特征库。通过收集到的有雾和无雾的自然图像检测实验证明,上述矢量特征库可信,并且所提出的方法具有良好的雾图检测能力,能为智能去雾提供很好的前置条件和方法。This paper briefly describes a technique which is applied to automatically separate hazing image from outdoor natural pictures. Image spectrum features are obtained by two - dimensional discrete Fourier transform. All these images get unified processing at RGB components and brightness, so that the correlation between image and the ambient light is the smallest. The gray - level co - occurrence matrix is applied to be regarded as features of classifi- cation. This paper collects a lot of outdoor color images, and adds a certain haze to the images by manual methods. Through training and evaluation, this paper builds vector feature database to separate fog image from all outdoor color images. It shows that vector feature database and the technique are credible. It proves that this technology can be used to predict better whether an image exists haze or not by the experiments of separating haze image and haze - free natural image.

关 键 词：支持向量机 频谱特征 灰度共生矩 雾图检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]