Hierarchical Dictionary Constructing Method for the Parametric Eigenspace Method

The parametric eigenspace method is an object recognition method based on visual learning approach with image coding technique. In this paper, to improve matching efficiency, a novel approach to construct hierarchical dictionary for the parametric eigenspace method is proposed. In the proposed constructing method, learning image set is classified hierarchically, and tree-structured dictionary is constructed from the classification results. For the learning image set obtained by varying one parameter value, the proposed method determines the class boundaries that are optimal in the sense of discrimination and least squares. Experimental results show that the tree-structured dictionaries made by the proposed method improve matching efficiency without a decrease in recognition accuracy.