论文部分内容阅读
在基于距离像的识别中,双谱和零相位谱都是保留了相位信息的平移不变特征,但零相位谱的特征维数要远低于双谱,而识别性能也低于双谱。经过分析指出双谱识别性能优于零相位谱的主要原因在于频谱的幅度通过相乘得到了加权;然后给出了三种特征加权方法:幂法、频谱幅度加权以及基于Fisher判决率的加权方法;最后将Fisher判决率加权和幂变换法结合起来对零相位谱进行加权,获得了识别性能优于双谱,而特征维数远低于双谱的加权零相位谱,基于仿真数据的试验验证了结论的正确性。
In the distance-based image recognition, bispectrum and zero-phase spectrum are all translational invariant features that retain the phase information. However, the feature dimension of zero-phase spectrum is much lower than that of bispectrum and the recognition performance is also lower than bispectrum. After analyzing, the main reason that bispectrum recognition performance is better than zero phase spectrum is that the amplitude of the spectrum is weighted by multiplication. Then, three feature weighting methods are given: power method, spectral amplitude weighting and weighting method based on Fisher decision rate Finally, Fisher’s decision-rate and power-transform methods are combined to weight the zero-phase spectrum. The recognition performance is better than bispectrum, and the feature dimension is much lower than the bispectrum weighted zero-phase spectrum. Experimental verification based on simulation data The correctness of the conclusion.