An accurate approach for the computation of polar harmonic transforms

Abstract

Polar Harmonic Transforms (PHTs) can be used as a feature extraction technique by exploiting their orthogonal basis that generate a set of rotation invariant features. PHTs have distinct traits as compared with other transforms and moments. This is because of their less redundant information. Applied to digital images polar harmonic transforms suffer from various numerical and statistical errors. In this paper a computational framework based on two different numerical integration schemes which reduce the numerical errors in the computation of the transforms. Extensive experiments are carried out in order to assess the accuracy of the proposed technique in face recognition problem. Two standard face databases are employed, i.e., ORL and JAFFE, which contain images with different variations. Furthermore, the rotation invariant trait of the proposed approach is evaluated using different rotation angles. The proposed approach attenuates the numerical errors as well as provides numerical stability to PHTs of high orders.