一种改进的伪Zernike矩快速计算方法

An Improved Approach to Fast Computation of Pseudo-Zernike Moments

为了有效地利用伪Zernike矩进行图像分析和模式识别,针对传统伪Zernike矩快速计算方法在计算伪Zernike矩时复杂度大的问题,提出一种改进的伪Zernike矩快速计算方法.该方法利用C lenshaw递推公式实现了伪Zernike矩多项式求和的快速计算.初步实验结果表明:在计算指定阶伪Zernike矩时,文中方法比传统伪Zernike矩快速计算方法需要更少的CPU时间;在人脸特征的提取及识别方面,文中方法的识别率比传统的主成分分析方法约高5%,而特征提取需要的平均时间为1.2 s.

In order to effectively apply pseudo-Zernike moments to the image analysis and the pattern recognition, an improved approach to the fast computation of pseudo-Zernike moments, which uses Clenshaw recurrence formula to implement the fast computation of pseudo-Zernike moment polynomial summation, is proposed to reduce the computational complexity of the traditional approaches. Preliminary experimental results demonstrate that the proposed approach costs less CPU time than the traditional methods in terms of computing special-order moments, and that it is of a recognition rate 5% more than that of the traditional principal component analysis method in terms of extracting and recognizing face features, with an average extraction time of 1.2 s.