Geogram based feature description
Gwangmin Choe, Tianjiang Wang · 2013
We introduce the concept of a geogram which is a generalization of spatiogram that includes more rich features. The spatiograms contain some moments upon the coordinates of the pixels corresponding to each histogram bin while the geograms contain information about the perimeter of grouped regions in addition to features in spatiogram. The perimeter of certain region has capability to represent the geometrical compactness of a feature distribution. The geogram based feature descriptor increases accuracy of tracking because of the descriptor using features in lower levels. Also, we consider that a convergence process of mean shift based on spatiogram is divided into obvious dynamic and steady states as well as in the control system, and introduce a technique in automatic control region, or a hybrid technique of geogram and traditional histogram, into controlling of convergence process. Moreover, we derive a formula for mean shift based on the proposed geogram and analyze the convergence behavior of this algorithm.