pi-SIFT: A Photometric and Scale Invariant Feature Transform
Jae‐Han Park, Kyung-Wook Park, Seung‐Ho Baeg, Moon-Hong Baeg · InTech eBooks · 2010
Pattern Recognition, Recent Advances 138In this paper, we propose a novel Photometric quasi-Invariant SIFT (PI-SIFT) describing features that are both invariant to geometric and photometric variations.In order to induce photometric quasi-invariant features, we first use the dichromatic reflection model (S. A. Shafer, 1985) which describes the light reflected at the material surface and the light reflected from the material body.The spatial derivative of this model, which gives the photometric derivative structure of the image, links differential-based features such as edge and corner to the theory of photometric invariance.Next, in order to obtain the features that are invariant to geometric variations such as translation, rotation, and scaling, we build scale-spaces based on the photometric quasi-invariant features.Finally, The same strategy of SIFT (D.G. Lowe, 2004) is used to build key-point descriptors.