Share Email Print

Proceedings Paper

The optimal extraction of feature algorithm based on KAZE
Author(s): Zheyi Yao; Guohua Gu; Weixian Qian; Pengcheng Wang
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

As a novel method of 2D features extraction algorithm over the nonlinear scale space, KAZE provide a special method. However, the computation of nonlinear scale space and the construction of KAZE feature vectors are more expensive than the SIFT and SURF significantly. In this paper, the given image is used to build the nonlinear space up to a maximum evolution time through the efficient Additive Operator Splitting (AOS) techniques and the variable conductance diffusion. Changing the parameter can improve the construction of nonlinear scale space and simplify the image conductivities for each dimension space, with the predigest computation. Then, the detection for points of interest can exhibit a maxima of the scale-normalized determinant with the Hessian response in the nonlinear scale space. At the same time, the detection of feature vectors is optimized by the Wavelet Transform method, which can avoid the second Gaussian smoothing in the KAZE Features and cut down the complexity of the algorithm distinctly in the building and describing vectors steps. In this way, the dominant orientation is obtained, similar to SURF, by summing the responses within a sliding circle segment covering an angle of π/3 in the circular area of radius 6σ with a sampling step of size σ one by one. Finally, the extraction in the multidimensional patch at the given scale, centered over the points of interest and rotated to align its dominant orientation to a canonical direction, is able to simplify the description of feature by reducing the description dimensions, just as the PCA-SIFT method. Even though the features are somewhat more expensive to compute than SIFT due to the construction of nonlinear scale space, but compared to SURF, the result revels a step forward in performance in detection, description and application against the previous ways by the following contrast experiments.

Paper Details

Date Published: 15 October 2015
PDF: 8 pages
Proc. SPIE 9674, AOPC 2015: Optical and Optoelectronic Sensing and Imaging Technology, 967438 (15 October 2015); doi: 10.1117/12.2204977
Show Author Affiliations
Zheyi Yao, Nanjing Univ. of Science and Technology (China)
Guohua Gu, Nanjing Univ. of Science and Technology (China)
Weixian Qian, Nanjing Univ. of Science and Technology (China)
Pengcheng Wang, Nanjing Univ. of Science and Technology (China)

Published in SPIE Proceedings Vol. 9674:
AOPC 2015: Optical and Optoelectronic Sensing and Imaging Technology
Haimei Gong; Nanjian Wu; Yang Ni; Weibiao Chen; Jin Lu, Editor(s)

© SPIE. Terms of Use
Back to Top