Visualization of object image database by using logistic discriminant analysis

For object detection, large-scale databases obtained via an in-vehicle camera have been proposed. The databases are generally used to evaluate object detection methods and/or to train classifiers in these methods. When proposing a new database, we should evaluate the characteristics of a large number of the samples in the database to improve the usability of the proposed database. In the evaluation, discriminant analysis is useful for visualizing the sample distribution of the database, and the visualized distributions show the statistical characteristics of the annotated (labeled) samples. In this paper, we propose a database analysis procedure using logistic discriminant analysis and demonstrate the visualization and evaluation of an object image database.