RETOP: A retinal topography keypoint descriptor
Jiung-Yao Huang, Hung-Ya Tsai, Chung-Hsien Tsai · 2016
Most visual applications rely on matching keypoints between two images. Due to the physical constraints of embedded devices, a descriptor with high accuracy and low memory is preferred for visual applications on mobile devices. Fast Retina Keypoint (FREAK) are known as the fastest descriptor nowadays with robustness on scale, rotation and noise. Based on the sampling points of FREAK, this paper proposes a new descriptor, RETOP, that has similar matching effectiveness as FREAK with three times faster speed. RETOP is designed according to the topography of ganglion cell density map of human retina and earn its name accordingly. Benefit from RETOP's well symmetrical structure, our experiment shows that RETOP is more efficiency than FREAK in the change of environments. Hence, RETOP make the realization of real-time visual application on mobile device on step further.