Two-Dimensional Extended Feature Line Space for Feature Extraction

A novel matrix-based image feature extraction approach, entitled Two-Dimensional Extended Feature Line Space (2DEFLS), is proposed in this paper. Nearest feature line (NFL) is a powerful classifier. Some NFL based subspace algorithms have been put forward recently. In most of the NFL-based subspace learning algorithms, the input samples should be vectors. For image classification tasks, image samples should be transformed to vectors firstly. This process leads to a high computational complexity and also may result in the loss of the geometric feature of samples. The proposed 2DEFLS is a matrix-based algorithm. It aims to minimize the within class scatter of an extended training sample set based on two-dimensional NFL. The experimental results demonstrate the effectiveness of the proposed algorithm.