一种改进的伪Zernike矩的快速算法研究

Zernike矩因具有正交性和旋转、平移、尺度不变性被广泛应用于模式识别特别是图像分析领域. 伪Zernike矩与Zernike矩相比, 具有更多的特征向量以及更强的抗噪声能力. 但在求伪Zernike矩时, 计算复杂性很大. 针对该问题, 在Chong和Xia等人对伪Zernike多项式的研究基础上, 结合Clenshaw递推公式, 提出了一种改进的快速求解算法. 实验结果表明, 改进后的方法在CPU的时间耗费上有较大的改善, 并初步将其应用于人脸识别中, 在训练和识别的时间上, 获得了较好的效果.

An Improved Algorithm Research for Fast Computation of pseudo-Zernike Moments

Zernike moments are widely used in the field of pattern recognition, especially image analysis, due to their good properties such as orthogonality and rotation, shift, scale invariance. Pseudo-Zernike moments have similar properties as those of Zernike moments; furthermore, they provide more feature vectors and are less sensitive to image noises than Zernike moments. However, the computational complexity of solution for the pseudo-Zernike moments needs much expensive time. Chong and Xia et al. have proposed fast algorithms respectively after they studied the pseudo-Zernike polynomials. But the problem of computational complexity has not been well solved. In this paper, addressing the problem, an improved fast computation approach integrating Clenshaw recursive formula is proposed. Experimental results show that the improved method is less computationally expensive. In addition, we preliminarily use pseudo-Zernike moments to face recognition and obtain preferable results about consumption of training and recognition time.