Principal Component Analysis of Point Distance Histogram for Recognition of Stamp Silhouettes

The paper presents a problem of stamp shape recognition. A stamp is given as a bitmap containing binary values, and may be represented by a specific geometrical form coming from the tradition of stamping process, which includes round, oval, square, rectangular or triangular shapes. While the problem of stamp detection, localization and extraction was addressed in several previous publications, in this paper we deal with the stage of features extraction and reduction, by means of Point Distance Histogram (at the stage of features extraction) and Principal Component Analysis (at the stage of dimensionality reduction). The final classification employs similarity evaluation involving hand-drawn templates, ideal shapes and average descriptors calculated for the entire database. The paper provides also some experimental results on real documents with different types of stamps and a comparison with a classical PCA applied on image matrix.