Region-based shape descriptor invariant to rotation, scale and translation

摘要: Abstract A region-based shape descriptor invariant to rotation, scale and translation is presented in this paper. For a given binary shape, positions of pixels belonging the are regarded as observed vectors 2-D random vector two eigenvectors obtained from covariance matrix population. The divided into four sub-regions by principal axes corresponding at center mass shape. Each sub-region subdivided same way. sub-division process repeated for predetermined number times. quadtree representation with its nodes regions derived above process. Four parameters translation, rotation calculated region each node while extracted root node. represented all similarity distance between shapes summing up absolute differences element vectors. Experimental results conforming MPEG-7 core experiments presented.