作者: Andrei Bursuc , Giorgos Tolias , Hervé Jégou
关键词:
摘要: In this work we design a kernelized local feature descriptor and propose matching scheme for aligning patches quickly automatically. We analyze the SIFT from kernel view identify reproduce some of its underlying benefits. overcome quantization artifacts by encoding pixel attributes in continous manner via explicit maps. Experiments performed on patch dataset Brown et al. [3] show superiority our over methods based supervised learning.