Comparative study of global invariant descriptors for object recognition

Even if lots of object invariant descriptors have been proposed in the literature, putting them into practice in order to obtain a robust recognition system face to several perturbations is still a studied problem. After presenting the most commonly used global invariant descriptors, a comparative study permits to show their ability to discriminate between objects with few training. The COIL-100 image database that presents a same object translated, rotated and scaled is used to test the invariance face to geometrical transforms. Partial object occultation or presence of complex background are examples of used images in order to test the robustness of the studied descriptors. The article compares them in a global and in a local context (computed on the neighborhood of a pixel). The SIFT descriptor is used as reference for local invariant descriptors. This study shows the relative performance of invariant descriptors used in a global and in a local context and identifies the different situations they are best suited.