Block Splitting Type Morphological Associative Memory and Its Recall Rate Improvement by Majority Logic Approach

A block splitting type MAM (BMAM) is one of the associative memories. The BMAM is a kind of the MAM without kernel images and is superior to other ordinary MAMs without kernel images in terms of the recall rate and the memory size. However, the perfect recall rate of the BMAM is inferior to that of the MAM using the kernel image. In order to improve the perfect recall rate, we proposed a majority logic approach for the BMAM. We paid attention to the BMAM’s features: multiple outputs can be obtained from single image by using different split manners and most units contained in common to the stored pattern appear in these output. In this paper, we confirm the performance of the BMAM employing the majority logic approach by multiple autoassociation experiments and discuss the effective usage considering the memory size.