A Graded Approach to Shape Representation

Shape representation is one of the key issues in a computer vision system. One important group of shape representations is multiple scale representation. In past years, many studies have been taken focusing on such group. Most of them used some operations such as smoothing, bluring or grouping to gradually obtain representation with diierent degrees of detail. Based on these operations, they used some parameters of their operations to indicate the detailing degree. One defect of using operation parameter is that human cannot easily make sense how detail the obtain representation is. To overcome such defect, we try to propose a graded approach which is guided by a human given similarity requirement to construct a most concise description of the shape which satisses the given requirement. As a result, our shape representation catch human's intuition about the detailing degree and suits machine's requirement for the concise description. c i ith cluster in cluster record C. c i index of starting vector of cluster c i. c i index of the middle vector of cluster c i. d ki the straight distance from x k to the t i. improvement of an reassignment from u to u (new). N S the number of the members of set S. t i the set of vectors that support the token t i. Q i a partial contour. the range of closeness for merge procedure. S the intersections of the contour and the skeleton. s i the ith vector in the set S. the required value of the similarity ratio. T the similarity ratio of T. T an approximation of a shape, i.e., a set of tokens. t i the token that corresponds to c i. 3 ~ c t i an attribute, center vector, of token t i. ~ o t i an attribute, orientation, of token t i. ~ s t i an attribute, scale, of token t i. ~ d t i the distance square error of token t i. ~ t i the similarity ratio of token t i. u(x k ; t i) the membership degree function of a vector x k with respect to token t i. " a given distortion threshold in token-detection procedure.