基于分形和纹理特征的红外图像云检测

Research on Infrared Image Cloud Detection Based on Fractal and Texture Features

针对红外图像中的云检测问题,采用了一种分形维数+纹理特征的检测方法。通过计算红外图像的平均分形维数,并根据阈值正确预判图像是复杂地形还是云区域,解决了无云图中的云的分类错误问题。通过构造局部云图像的灰度共生矩阵,得到反映图像纹理特征的四个二次统计特征,云区域和无云区域在四维特征上具有明显的区分度,根据该特征训练的非线性SVM分类器能够有效地区分云区域和无云区域。通过分形维数预判以及由统计特征训练的SVM分类器,可实现云区域的精确检测与标记。上述方法对仿真、真实红外图像进行了验证,具有实用性,准确率有较大提高。

Aiming at the problem of cloud detection in infrared image,we proposed a detection method based on fractal dimension and texture feature.By calculating the average fractal dimension of the whole infrared image,the method correctly predicts whether there is a cloud region in the image according to the threshold value,and solves the problem of classification error of cloud in cloudless image.By constructing the Gray-level co-occurrence matrix(GLCM)of local cloud images,the method can obtain four statistical features that reflect the texture features,which has obvious discrimination in the four-dimensional feature space.By the prediction using fractal dimension and SVM classifier trained with statistical features,the method achieves that cloud region in infrared image can be detected and marked precisely.Then detailed experiments on simulation and real images are also given to verify the correctness and effectiveness of algorithm in cloud detection applications.The method is proved to achieve a high accuracy,and has its applicability.