An Effective Blur Invariant Interest Point Detector

Abstract

In order to extract interest point under some object deformations, such as image blur, geometric deformation et al., an invariant interest point detector based on Gabor multi scale-space was proposed. Firstly, build the Gabor multi scale-space representation by smoothing the image with a series of Gabor filters, secondly, detect the interest points by searching maximum response in the Gabor multi scale-space and determine the characteristic scale directly, finally, describe the local structure of interest point. The experimental data demonstrate that the proposed detector obtains the best performance under image blur in terms of repeatability and matching score.