Recognizing Human Faces with Tilt

Issues related to realtime face recognition are perpetual even with many existing approaches. Generalizing these issues is tedious over different applications. In this paper, the real time issues such as tilt or rotation variation and few samples problem for face recognition are addressed and proposed an efficient method. In preprocessing, an edge detection method using Robert's operator is utilized to identify facial borders for cropping purpose. The query images are axially tilted for different degrees of rotation. Both database and test images are segmented into one hundred fragments of 5 * 5 size each. Four different matrix characteristics are derived for each divided part of the image. Corresponding attributes are added to yield features related to final matrix. Final one hundred facial attributes are obtained by fusing diagonal features with one hundred features of matrix. Euclidean distance between the final attributes of gallery and query images is computed. The results on Yale dataset has superior performance compared to the existing different approaches and it is convincing over the dataset created.