Eigenvalue Analysis with Hough Transform for Shape Representation and Classification

In this work, we present eigenvalue based shape descriptor (EHough) which makes use of small eigenvalue and large eigenvalue along with Hough Transform to obtain the dominant features. The small eigenvalue and large eigenvalue are computed for each pixel associated with a shape boundary. In order to compute eigenvalues, we have taken every pixel associated with a shape boundary and its connected pixels within a window of certain size. Each pixel under processing is replaced by these eigenvalues which results in two matrices. These two matrices capture the structure of a shape. It is well known fact that the Hough transform is region based and is well suited under noise conditions. Hence, we perform Hough Transformation on these two eigenvalue based matrices to obtain compact representation of the shape and these features are matched using Euclidean Distance. We have performed decision level fusion of proposed approach with blockwise binary pattern (BBP) to enhance the classifier accuracy. Extensive experimental results on the publicly available shape databases namely, Kimia_99 and Kimia_216 and MPEG_7 data sets demonstrate the accuracy of the proposed method. The results of the experiments exhibit the success of proposed approach, in comparison with well-known algorithms from the literature.