The Study Of Color Pattern Recognition Based On The Joint Fractional Fourier Transform Correlator

Optical correlation systems have been successfully developed for character recognition and target detection.There are two common correlation systems,matched filtering correlator(MFC) and joint transform correlator(JTC).As an important real-time image processing system,JTC is applied to the remote sensing,the medicine, the military,and the industrial automatic check and so on.Combined with the system of multi-channel double light path and JTC,we propose a new system of multi-channel double light path joint fractional Fourier transform correlator.The theoretical analysis and optical experiment have been performed,otherwise,experiment result obtained from the new system has been compared with that from the conventional system,which shows the new system has a good ability of recognition.Moreover,rotation-invariant color pattern recognition based on the new system has also been performed.We have mainly investigated three aspects as follows:First,we propose a new system of multi-channel double light path joint fractional Fourier transform correlator for color pattern recognition.Compared with the conventional JTC based on which,one can obtain 36 correlation expressions which form 15 locations of correlation peaks at the output plane,due to the shift-variant property of fractional Fourier transform(FRT),the fractional spectra of three channels can be separated at the frequency plane,that is to say,R_R,R_G,R_B are absolutely separated and T_R,T_G,T_B are absolutely separated.On the other hand,R_R and T_R,R_G and T_G,R_B and T_B can completely overlapped at the frequency plane owing to the double light path system,we can obtain three correlation peaks at the output plane which show a pair of cross-correlation peaks that is desired and one autocorrelation peak,so the noise is greatly reduced,which optimize the conventional system.The theoretical analysis and optical experiment are discussed in detail.Experimental results are presented to demonstrate the ability of this system.Second,based on the proposed system,we have performed rotationa-invariant color pattern recognition,in which the synthetic discriminate function(SDF) is quoted at the input plane to store the reference images.To optimize the correlation peaks,the joint power spectrum is binarized to perform Fourier transform.Compared with the result obtained from the non-binarized joint fractional power spectrum,it can be obviously found that the result obtained from the binarized joint fractional power spectrum is reasonably better,which has less sidelobes and furthermore,the correlation intensity peaks are sharper,hence increasing its discrimination ability.Third,to recognize the exact rotation angle of the target image,we have made a set of rotational color patterns with different in-plane rotation angles at 1°interval between 0°and 359°.Based on the theory of probability,the angle region of valued rotational gesture has been reduced until the target image has been recognized.The response time of SLM and CCD is 25ms and 30frames/s respectively.The optical experiment cost only a few seconds,which shows the novel system has a good ability of real-time recognition.