Pattern location estimation for multichannel images

Abstract

We propose pattern location algorithms for multichannel images. Colored images are taken as example for simulations and experimental results. We consider the model of a target with different luminance in each channel, and Gaussian additive noise, independent and with different noise levels in each channel. The maximum likelihood (ML) estimation of the location of the target in a given scene leads to optimal combination of the multichannel information. We develop algorithms to obtain the target location when the target luminance and the variance of the noise are either known or unknown in each channel. The proposed algorithms have a total complexity of a correlation. A comparison with the results obtained by classical addition of correlations values calculated in each channel is made, demonstrating the superiority of the optimal algorithms. Simulations are developed to test the robustness of the proposed algorithms when the hypothesis of Gaussian and white noise are not fulfilled. The results show that the performance of the estimation do not reduce for exponential white noise or uniform white noise, but the performance deteriorates when the noise is correlated.