为了更好地兼顾特征描述子的鲁棒性和生成复杂度,提出了一种基于Laplacian的局部特征描述算法.分析说明了Laplacian不仅对图像的欧氏变换、缩放及亮度线性变化具有较好的性质,而且能够反映图像的局部结构特征.据此利用高斯型拉普拉斯变换响应建立了一种64维特征描述子,并将该特征描述子应用于特征点匹配.匹配实验结果表明,在图像尺度缩放、旋转、模糊、亮度变化和较小视角变化等多种变换条件下,该描述子不仅能够取得较好的匹配效果,而且匹配速度是尺度不变特征变换(SIFT)的4倍以上.该算法适用于实时性要求较高,存在旋转、尺度缩放、亮度差异、图像压缩变换以及视角变化不大的结构图像间的匹配.%In order to balance the robustness and building complexity of a feature descriptor, a local feature description algorithm based on Laplacian is presented. It is analyzed and illustrated that the Laplacian not only has good properties to Euclidian transformation, zoom, and linear brightness changes of an image, but also can characterize the local structure of the image. On the basis of that, a 64-dimension descriptor is built with the response of Laplacian of Gaussian. Finally, the descriptor is used to match feature points with the absolute distance as similarity measurement. Simulation results indicate that the proposed descriptor can obtain better matching results for the image with zoom, rotation, blurring, illumination varying as well as smaller viewpoint changes, and the matching speed is more than 4 times that of Scale Invariable Feature Transformation (SIFT). The proposed feature description algorithm is suitable for matching the images of structured scenes,for it is insensitive to the image transformation with rotation, zoom, luminance varying, compression or small viewpoint changes.
展开▼
