Image Recognition Using Weighted Two-Dimensional Maximum Margin Criterion

In image recognition, feature extraction techniques are widely used to enhance discriminatory performance. In this paper, a new method for image feature extraction, called weighted two-dimensional maximum margin criterion (W2DMMC), is proposed. Different from conventional maximum margin criterion (MMC), W2DMMC is directly based on two-dimensional image matrix rather than one-dimensional vector. And W2DMMC has an additional weighted parameter beta that further broadens the margin. W2DMMC completely circumvents the small sample size problem and is computationally efficient. As a connection to 2DLDA, we show that 2DLDA can be recovered from W2DMMC when imposing some constraints. The better performance of W2DMMC in terms of both recognition accuracy and training time is demonstrated by experiments on real data set.