dc.description.abstract | Abstract
Logarithmic harmonic filter can detect objects at different projection angles, while Mellin radial harmonic filter has the ability to detect objects of different scales. The Mexican-hat wavelet function can extract edges of equal width for objects, regardless of their sizes and orientations. Hence, incorporation of wavelet filtering in the logarithmic harmonic and Mellin radial harmonic filter, the LHW and MRHW filter, which is also modified to eliminate the pre-processing, can improve the pattern recognition performance as compared with the LH and MRHW filter. The theory is presented together with computer simulation.
The LHW filter is capable of identifying the input object, within a wide range of projective angle, , with a variance under 30%. On the other hand, the MRHW filter can distinguish the input object with a wide allowable scale range 1~0.25. The correlation peaks are sharp and the peak intensity is relatively uniform with a variance under 30%. Computer simulation is adopted to investigate the performance of the LHW and the MRHW filter. Experimental results, including those obtained under the white noise, implemented in the Joint-transform correlator are also presented.
In the prior research, generally, collimated wave is used both as the incident wave and as the readout wave in a joint transform correlator. Here, we propose the use of converging wave instead to read out the joint transform spectrum, which is recorded in a thick recording medium. The recording medium is conceptually divided into numerous thin layers. Assume that the incident readout wave is scattered only once by each layer and the directly transmitted wave is not significantly attenuated. We arrive at the theory for low diffraction efficiency limit, which is then numerically tested. The result shows that the correlation peak intensity (CPI) is the maximum when the radius of curvature of the readout wavefront is the same as the focal length of the Fourier transforming lens. However, the CPI is quite independent of the radius of curvature of the readout wave under normal experimental conditions. Minimum thickness for the recording medium beyond which the CPI doesn’t increase significantly is obtained. In the condition that requires good discrimination among similar objects and precise determination of the object position under noisy environment, the CPI depends strongly on the curvature of the readout wavefront, particularly for the portable correlator which requires short focal length for the Fourier transforming lens. | en_US |