二元微分联合变换相关器实现图像识别的研究

Study about the binary differential joint transform correlator used in pattern recognition

为了克服二元联合变换相关器的取阈和实时性问题,采用微分运算与二元联合变换相关器结合的方法。联合功率谱中,由于干扰项随频率的变化率比需要的信息量随频率的变化率慢得多,因此可以采用微分方式削弱干扰项,增强互相关信号,对微分后的功率谱再进行二值化非线性处理得到优化功率谱,然后对优化功率谱进行傅里叶逆变换,得到目标和图像的相关输出。模拟实验结果表明,所提出的相关器能很好地抑制自相关峰,增强互相关峰,具有更强的鲁棒性。而且该方法在处理过程中,不需选择阈值,简单的微分及二值化操作使之具有更强的实时性。

New binary joint transform correlator （BJTC） based on differential operation was proposed. It can well resolve the disadvantages about the bottleneck of real-time processing and thresholding selection in BJTC. The unwanted term in joint power spectrum （JPS） varies slowly compared with wanted term. It is reasonable to use differential operation to improve JTC＇s contrast by depressing the auto-correlate peak and enhancing the cross-correlate peak. Then the differential JPS was binarized to only two values at Fourier plane before inverse Fourier transform. Then higher cross-correlate output between references and object image can be obtained. Simulation results show the proposed JTC can well weakens auto-correlation peak and enhances cross-correlation peak. It also has better robustness. The proposed method can avoid computing or pre-setup threshold. Furthermore, it＇s quite simple differential operation and binarization require far fewer digital computations than does the BJTC.