ABSTRACT
In this paper, the above-mentioned problem is firstly dealt with. We adopt modified Zernike moments to define local distribution similarity. The distribution feature of a region is denoted as a vector which is composed by a series of modified Zernike moments with different repetitions and orders. We also provide a quadtree based approach in order to speedup the very time-consuming search on spatial map. We partition the whole map into many cells with identical width and length, and then a quadtree is employed to hierarchically cluster all the cells. To achieve efficient searching, category based pruning and PCA based pruning is adopted to eliminate as many dissimilar clusters as possible.
