Validating directional edge-based image feature representations in face recognition by spatial correlation-based clustering

Directional edge-based image feature representations have already been developed and applied to medical radiograph analysis, face detection, and face identification. The scheme of the feature-vector generation is considered as a dimensionality reduction from a high-dimension edge feature map space to a low-dimension edge histogram space. However, the validity of such schemes in dimensionality reduction has not yet been studied on the sound basis of statistics. In this paper, in order to validate the directional edge-based feature representations, the scheme of dimensionality reduction employing the hierarchical clustering based on the spatial correlations of edges has been investigated. In this study, the feature representations of images have been evaluated using facial images as the test vehicle. As a result, the validity of using the directional edge-based feature vectors in image recognition has been verified by the similarity between the results of the hierarchical clustering and the schemes employed in the feature representations.