Design of matched filters for image correlation recognition of distortion targets

A matched filter with great tolerance of distortion was designed by using the optimal tradeoff synthetic discriminant function to overcome the low discrimination rate in the correlation recognition of the rotating and rescaling targets. The matched filter with the bandpass feature was obtained by adjusting the contribution of output noise, average power spectral density of training sample image and training sample images′ similarity to the average value of training sample images. The simulation experiment of the correlation recognition for the target with rotation angle of 0～30° and the variation dimension range of 0～20% with a single matched filter was carried out. The designed filter resembles a band-pass filter. Simulation results demonstrate the designed filter has sharp output peaks, good discrimination and high tolerance for distortion.