DefinePK

Abstract

Face Recognition using images obtained from the unconstrained environment is the challenge, yet to be resolved. This situation is due to cluttered background and poor lighting conditions or illumination. Capturing images from a long distance, atmospheric turbulence, out of focal length and camera in motion are also the reasons behind the drastic decline in the performance of face recognition. A novel three-step formula has been proposed in this paper to address the issues from existing methodologies and provide the consistent accuracy in final face recognition. First and foremost, the query image of a face is thoroughly analyzed to know the blur presence and its type. Later, the model images are also blurred to the same extent as of query image and face recognition is done using deblurring both model and query images by Iterative Graph based image restoration technique. The accuracy of the face recognition using the proposed algorithm is consistent under different levels and types of blurring. The performance of the proposed method (for blurring and deblurring the models) is validated for different classification algorithms namely Collaborative Representation Classification (CRC), Relaxed Collaborative Representation (RCR) and Linear Collaborative Discriminant Regression Classification (LCDRC). LCDRC outperformed the existing peers in accuracy and robustness. The best recognition rate of 96.25 % is obtained for blur face images using this proposed method.