Real time hybrid target recognition using generalized and wavelet pre-processed matched filters

One special type of complex matched filter (CMF), namely, amplitude-modulated phase-only filter (AMPOF) effectively utilizes both amplitude and phase information to obtain a narrower and larger autocorrelation peak. The AMPOF is optimized for improved target recognition applications. Both correlation performance and output signal-to-noise-ratio (SNR) for this optimized filter are found to be enhanced within the theoretically identified range of the filter parameters. The real-time implementation issues for the CMFs are addressed next. To assure real-time automatic target recognition (ATR), we develop a generalized expression for the family of CMFs. In particular, both amplitude and phase of the AMPOF are ternerized to yield an improved performance of the correlator. In order to achieve rotation-invariance for the CMFs, we proposed an amplitude-coupled minimum-average-correlation-energy (MACE) filter. This amplitude-coupled MACE filter is implemented and tested for its rotation invariance feature. The algorithm is also investigated for probability of correct classification using synthetic aperture radar (SAR) images.; The relatively new but powerful tool, namely, wavelet transform (WT) can also be utilized appropriately in the processing of CMF. Accordingly a WT pre-processed AMPOF is introduced and implemented next. It is evaluated in terms of various target recognition performance statistics. The effect of nonlinear thresholding in suppressing noise has also been studied. The WT pre-processed and nonlinear thresholded AMPOF is found to yield significantly better performance than the AMPOF. Finally, a hybrid character recognition system that uses a feature extraction method is proposed. The features are extracted using WT, pre-classified using a k-nearest neighbor neural network and subsequently post-processed using an optical correlator. This feature-based neural wavelet optical architecture is finally tested on blurred character images.