Selecting feature lines in generalized dissimilarity representations for pattern recognition

Recently, generalized dissimilarity representations have shown their potential for small sample size problems. In generalizations by feature lines, instead of dissimilarities with objects, we have dissimilarities with feature lines. One drawback of such generalization is the high amount of generated lines that increases computational costs and may provide redundant information. To overcome this, the selection of lines based on the length of the line segments has been considered in previous works, showing good results for correlated data. In this paper, we propose a new supervised criterion for the selection of feature lines. Experimental results show that the proposed criterion obtains competitive or better results than those obtained by previous criteria, especially for data with high intrinsic dimension, spherical data and data with outliers. As our proposal provides better results for small representation sets, it allows one to obtain a good trade-off between classification accuracy and computational efficiency.