Research On Morphological Hetero-Associative Memory Based On Feature Of Image

Morphological Hetero-associative Memories (MHM) is a kind of simulation mathematical model of associative memory function of human brain. Compared with the traditional associative memory model, MHM not only has capability of handling binary pattern but also can deal with the real mode, and it has good noise-tolerance to single erosive or dilative noise, as well as broad application prospects and strong vitality in pattern recognition, image processing, and so on. Although MHM has many advantages, it can’t provide perfect recall even if there is no noise in the input pattern. For the shortcoming of MHM, the following work is done in this paper.Firstly, the MHM of the binary pattern has been studied, and proposed the theorem2. The theorem gives a sufficient condition about MHM of binary pattern can provide perfect recall. According to this condition, this paper used the binary image feature information respectively to modify the input mode of W memory algorithms and M memory algorithm and proposed two new algorithms-BFW algorithm and BFM algorithm. Finally, the two algorithms have been proved that they have better memory performance than MHM by using the examples.Secondly, the MHM of the real pattern has been studied, and proposed the theorem3. The theorem gives a sufficient and necessary conditions about MHM of real pattern can provide perfect recall. According to this condition, this paper used the gray-scale image feature information respectively to modify the input mode of W memory algorithms and M memory algorithm and proposed two new algorithms-GFW algorithm and GFM algorithm. Finally, the two algorithms have been proved that they have better memory performance than MHM by using the examples.At last, in this paper, a novel secret image sharing algorithm was proposed based on GFW algorithm or GFM algorithm. This algorithm can get the memory matix by GFW or GFM algorithm recalling to false image and secret image. And then we add the random single-noise to the false image and memory matrix, so we can get n false images and n memory matrices which contain noise and distribute them to n participants. But only n-1participants can get n-1no-noise images through the n-1noise images which maximizing or minimizing. In the same way, n-1participants can get n-1memory matrices. We use no-noise false image as input image and use the no-noise memory metrix to get the secret image. Finally, the feasibility of the algorithm is varified by a practical example.