Research On Framework Generalization Of Morphological Associative Memories

Associative memory (AM) is one of important functions of the human brains, and also the source oflogical thinking and imaginary thinking, reasoning and innovation. Compared with traditional associativememory, Morphological Associative Memories (MAM) has great improvement in the memory performance.MAM not only has capability of handling binary pattern but also can deal with the real mode. Furthermore,it has the unlimited storage capacity, one-shot recall speed and good noise-tolerance to single erosive ordilative noise, as well as broad application prospects and strong vitality in image processing, patternrecognition, and so on. But it also has some problems, such as hetero-MAM can’t provide perfect recalleven if there is no noise in the input pattern. However, hetero-AM has much more widely application, so ithas important significance and value to improve the performance of hetero-AM. This paper digs into theframework of Morphological Associative Memories and presents new methods of AM, and then improvesthe performance of hetero-AM.Firstly, based on analyzing the norms and operators of MAM, the feature, which reverse operationsexist between memory and recall of the MAM, is pointed out. Through adopting reverse thinking, thereverse methods of MAM are proposed. The reverse methods and original methods are complementary inthe way of resisting noise. Therefore, these two kinds of associative memories methods can be combine toimprove the effect of hetero-associative and pattern recognition.Secondly, a new kind of Logarithmic and Exponential Morphological Associative Memories(LEMAM) is proposed, which based on studying the framework of MAM. Hetero-associative LEMAM cannot provide perfect recall. But in any cases, this method can recall perfectly or have much better result ofthe associative memories. Therefore, LEMAM and other morphological associative memories methods canbe combine to improve the effect of hetero-associative.At last, a new idea and method about grouping and partition is proposed against the defects ofhetero-MAM. The original input patterns are grouped firstly, and then the sets on model grouped arecalculated one by one in morphological methods. A local partial to global, and then the perfect recalls on morphological hetero-associative memories are achieved. This new idea and method can be widely appliedto pattern recognition and image processing.