Incorporating Gradient Magnitude in Computation of Edge Oriented Histogram Descriptor
Liangpeng Xu, Yong Li, Chunxiao Fan, Hongbin Jin, Xiang Shi · Electronic Imaging · 2016
This paper proposes an approach to employing the gradient magnitude in computing EOH descriptors. EOH has a better matching performance than SIFT (scale invariant feature transform) on multispectral images but does not utilize the gradient magnitude. In EOH, every edge pixel has the same contribution to the orientation histogram, which suppresses the usage of gradient magnitude. Observing this, we propose utilizing gradient magnitude with a logistic sigmoid function. The gradient magnitude of a pixel serves as the input to a sigmoid function, and the output is used as the weight of the pixel. Experimental results show that the proposed approach performs more robustly than the original EOH on multispectral images.