In this paper, we present a feature extraction method by utilizing an error estimation equation based on the Bhattacharyya distance. We propose to use classification errors in the transformed feature space, which are estimated using the error estimation equation, as a criterion for feature extraction. The construction of linear transformation for feature extraction is conducted using an iterative gradient descent algorithm, so that the estimated classification error is minimized. Due to the ability to predict error, it is possible to determine the minimum number of features required for classification. Experimental results show that the proposed feature extraction method compares favorably with conventional methods.
All Science Journal Classification (ASJC) codes
- Signal Processing
- Computer Vision and Pattern Recognition
- Artificial Intelligence