应用振幅调制纯相位滤波器实现旋转不变识别

Rotation invariant pattern recognition based on amplitude-modulated phase-only filter

提出了一种基于振幅调制纯相位滤波器(AMPOF)的旋转不变识别新算法。该算法充分利用振幅调制纯相位滤波器区分能力强的优点,结合综合鉴别函数方法,拓展了综合鉴别函数(SDF)在畸变不变模式识别领域的应用。利用各幅训练图像傅里叶变换的振幅和相位信息,逐一构建成子滤波器,并将它们加权求和得到合成滤波器;然后采用迭代算法求解出合理的权重系数。仿真实验结果表明:对于实验采用的目标训练图像和非训练图像,改进算法的相关输出峰值大约提高95%～112%;对于只有少量像素的局部训练图像,改进算法的相关输出峰值大约提高74%。这种优化算法使得振幅匹配和相位匹配更加充分,相关输出增益明显增强,能够一定程度地降低训练量。

In order to get good general performance of optical matched filtering correlation recognition, a new algorithm based on Amplitude-modulated Phase-only Filter（AMPOF） is proposed for rotation invariant pattern recognition. The algorithm extends the application of Synthetic Discriminant Funetion（SDF） to distortion invariant recognition efficiently by utilizing good discrimination capacity of the AMPOF and employs both amplitude and phase information of Fourier transforms of training images to construct a series of sub filters, and to get a synthetic filter by weighted sum of sub filters. Finally,the algorithm can obtain the weight coefficients by using a iterative algorithm. The simulation results indicate that the correlation peak value of improved algorithm can be enhanced by 95-112% for training images and non-training images in experiment, and can be enhanced by 74 % for one local training image with a few of pixels. The amplitude matching and phase matching will be much sufficient under new algorithm, consequently, the correlation output gain will be enhanced evidently, and the training data can be reduced to a certain extent.