基于无字典模型的红外与可见光图像融合分类

针对传统图像分类识别方法采用单一传感器获得图像存在的局限和不足，提出了一种基于红外和可见光融合的分类算法。首先分别对红外图像和可见光图像提取密集型尺度不变特征变换（D-SIFT），然后采用无字典模型（CLM）变换，并利用空间金字塔匹配（SPM）进行精细划分，最后用混合核支持向量机（SVM-CK）方式将红外和可见光在特征级融合并分类。在VAIS和RGB-NIR两个数据库上对该方法进行验证，融合后分类精度分别比单一图像源均有较大提高；与BoVW方法比较，精度分别提高了4.7%和12.1%。证明多数据源融合的方法综合了红外和可见光各自成像的优势，使获得的特征信息更完善，分类效果显著高于单一数据源的分类结果。

Infrared and visible image fusion classification based on a codebookless model (CLM)

In view of the shortcomings and limitations of traditional image classification algorithms based on a single image source, a novel image classification method based on infrared image and visible image fusion is proposed; this combines the advantages of infrared imaging and visible imaging, and thus makes the feature information much more abundant. In the first step, sets of dense scale invariant feature transform (D-SIFT) features were extracted from the infrared imaging and visible images. Then, an effective codebookless model (CLM) was employed, along with a spatial pyramid matching (SPM) strategy, which divides an image into regular regions. Finally, a composite kernel method based on support vector machines (SVM-CK) was utilized to fuse the extracted features and realize the final classification. The proposed method was validated using VAIS and RGB-NIR databases. The accuracy of fusion were both improved than that obtained using a single image source. Compared with traditional methods such as BoVW, the accuracy is improved by 4.7% and 12%, respectively. The experimental results show that the results of multidata fusion are significantly better than those based on single data sources.