A new method of evaluation and optimization for feature extraction

The feature extraction and selection is one of the core issues of pattern recognition. In this paper, taking the feature extraction and selection for the letters A-Z and Arabic numbers 0–9 as object of study, the feature extraction methods are evaluated by calculating the sum, the variance and the minimum distance of Euclidean distance while the feature dimensions are 32, 64, 96 and 128. The arrangement of feature column element data is optimized according by their variances, the separability measure values of 1–128 dimensional features are calculated, and the Mix combination feature optimization method of feature selection for each dimension is proposed.