Sparse Representation-based Classification of Farsi Handwritten Digits Using Fisher Discrimination Criterion and Local Linear Embedding (LLE)

摘要

In recent years, sparse representation-based methods have performed well in machine vision and image processing. The main challenge in designing a proper classifier to detect Farsi digits is modeling the data subspace and classification based on the presented model. In this paper, we propose a method based on sparse representation. Local Linear Embedding (LLE) to recognize the handwritten Farsi digits. The conducted approach to recognizing Farsi digits in this paper is sparse representation, based on Fisher Discrimination Criterion for linear transformation of the data. In this method, data sets belonging to the same class are grouped together, whereas data sets that belong to different classes are kept apart. LLE is used as a regulator to maintain the local neighborhood of the data. Experimental results of Hoda database test data with 80,000 samples indicate that the proposed method has a higher accuracy than previously presented methods and has achieved the accuracy of 99.36% for 60,000 training samples and 20,000 test data.