基于多尺度分析的稠密SIFT特征提取方法

Dense SIFT feature extraction based on multi-scale analysis

为改良稠密SIFT特征难以捕获图像深层细节信息的缺陷,提出一种基于多尺度分析的稠密SIFT特征提取方法。将小波分析的多尺度结构与稠密SIFT特征提取方法相结合,在小波空间中提取图像不同层面、不同方向上的稠密SIFT特征,通过特征融合生成的新特征保留稠密SIFT特征易于计算、采样点均衡等优点,兼具稀疏SIFT特征的多尺度结构,有利于发现图像中隐藏的深层信息。实验结果表明,该方法可有效提升图像分类结果准确率。

To improve the defect of the dense SIFT feature that it is difficult to capture the deep details of the image,a dense SIFT feature extraction based on multi-scale analysis method was proposed.The dense SIFT was combined with the wavelet transform,and the dense SIFT features in different level and different directions in the wavelet transform space were extracted.The new feature preserved the advantages of dense SIFT feature such as simple calculation and sampling point equalization.And also it had a sparse SIFT feature of multi-scale structure,which was helpful to find hidden information in the image.Experimental results show that the proposed method can effectively improve the accuracy of image classification results.