Understanding and Improving Kernel Local Descriptors
International Journal of Computer Vision(2018)
摘要
We propose a multiple-kernel local-patch descriptor based on efficient match kernels from pixel gradients. It combines two parametrizations of gradient position and direction, each parametrization provides robustness to a different type of patch mis-registration: polar parametrization for noise in the patch dominant orientation detection, Cartesian for imprecise location of the feature point. Combined with whitening of the descriptor space, that is learned with or without supervision, the performance is significantly improved. We analyze the effect of the whitening on patch similarity and demonstrate its semantic meaning. Our unsupervised variant is the best performing descriptor constructed without the need of labeled data. Despite the simplicity of the proposed descriptor, it competes well with deep learning approaches on a number of different tasks.
更多查看译文
关键词
Similar Patches,Efficient Match Kernel,Kernel Descriptors,Patch Maps,Hand-crafted Descriptors
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络