Recognizing faces from surgically altered face images using granular approach

This paper introduces a strategy for recognizing surgically altered face images using granular approach and similarity matching. In this method, global, local and granular features are used for recognizing the altered images. Global feature components are separated from the entire face pictures in the form of low frequency coefficients of Cosine change. For extracting neighborhood components, Gabor filtering is used. Granular features are also considered for the recognition purpose. Granular element implies, non-disjoint components extricated from various granular levels. When using the granulated data, more adaptability is attained in breaking down the data like nose, two eyes and mouth. These granular features are extricated using three levels of granularity. The primary level gives the global information about the face image. The second level gives the internal and external facial feature components. At the third level, highlights from the neighborhood facial locales are separated. These granular features are concatenated and a granular feature vector is formed. The final step is the recognition of face images by similarity matching.