(2D) ²UFFCA: Two-directional two-dimensional unsupervised feature extraction method with fuzzy clustering ability

In this paper, based on the principles of the maximum margin criterion (MMC) and by introducing the fuzzy method and the tensor theory into it, a novel matrix model fuzzy maximum margin criterion (MFMMC) is proposed. Also, on the basis of it, a two-directional two-dimensional unsupervised feature extraction method with fuzzy clustering ability ((2D) ²UFFCA) is constructed. This method can directly realize fuzzy clustering of matrix model data. And it can also achieve the two-directional two-dimensional feature extraction of them, that is, the realization of dimension reduction. At the same time, the adjusting parameter 7 in the matrix model fuzzy maximum margin criterion is defined reasonably from the respect of geometry intuition, which is proved theoretically. In order to improve the efficiency of feature extraction, an effective method which can find out the projection matrices of matrix model data is presented. The results of tests show the above advantages of the method.