Binary Shape Segmentation and Classification using Coordination Number (CN)*
We proposed a novel, micro-structures, shape-preserving local descriptor for contour-segmentation- based binary object classification and recognition, named local coordination number count (LCNC). In this method, we formulate the problem by estimating the 8-neighbourof each binary object pixel. In the matching stage, we used Euclidean distance between eigenvalues corresponding to correlation coefficient and the dynamic programming to find out the optimal correspondence between boundary smoothness of two shapes. Experimental results obtained from shape data bases demonstrate that the proposed LCNC can achieve better classification rates compared to existing shape descriptors.It produces detailed data on the distributed coordination numbers that relate to various types of contacts between small, medium, and large components. The emphasis of the study is on the mean coordination numbers associated with these contacts. These partial mean coordination numbers differ with the volume fractions of the components, according to the findings, while the overall mean coordination numbers vary with the volume fractions of the components while the overall mean coordination number is essentially a constant and independent of particle size distribution.