Manifold based Persian digit recognition using the modified locally linear embedding and linear discriminative analysis

Abstract

In this study, a new nonlinear manifold learning technique based on the Locally Linear Embedding (LLE) is proposed. In this method, a new modified LLE based on the neighborhood conception is proposed. Then, by this new definition of LLE, true neighbors of each data are selected to construct the reconstruction weights. By this new definition of neighborhood of each data, structure of data manifold is preserved in low dimensionality. In this study, after using the proposed MLLE, linear discrimination analysis (LDA) technique is applied on Persian handwritten character. Finally, recognition rate has been calculated by K nearest neighbor (KNN) classifier. Experimental results demonstrate the superiority of the proposed method.