Formulation of the MACE Filter as a Linear Associative Memory

The minimum average correlation energy (MACE) filter of [Mahal, Ravi], produces sharp correlation peaks over the recognition class in pattern recognition problems. It is known that the MACE filter of [Mahal, Ravi] can be formulated as a pre-whitening filter cascaded with a linear synthetic discriminant function. It will be shown that the MACE filter is equivalent to a linear associative memory (LAM) trained with pre-whitened patterns and as a result its coefficients can be computed iteratively via the LMS algorithm in the space domain. Experimental results will also be shown.