Anant Choksuriwong, Bruno Emile, H. Laurent, Christophe Rosenberger
Journal of Electronic Imaging, Vol. 17, Issue 02, 023015, (April 2008) https://doi.org/10.1117/1.2912071
TOPICS: Databases, Object recognition, Transform theory, Detection and tracking algorithms, Sensors, Feature extraction, Image classification, Image analysis, Intelligence systems, Zernike polynomials
Although many object invariant descriptors have been
proposed in the literature, putting them into practice to obtain a robust
recognition system that is able to face several perturbations is
still a studied problem. After presenting the most commonly used
global invariant descriptors, a comparative study permits us to show
their ability to discriminate between objects with little training. The
Columbia Object Image Library database (COIL-100), which presents
a same object translated, rotated, and scaled, is used to test
the invariant features of geometrical transforms. Partial object occultation
or presence of complex background are examples of used
images to test the robustness of the studied descriptors. We compare
them in both a global and a local context (computed on the
neighborhood of a pixel). The scale invariant feature transform descriptor
is used as a reference for local invariant descriptors. This
study shows the relative performance of invariant descriptors used
in both a global and a local context and identifies the different situations
for which they are best suited.