Homomorphic filtering and reflectance ratio based improved illumination invariant face recognition
What is it about?
HFRIN-SFDWT based is an efficient illumination invariant face recognition technique. We have introduced a new illumination normalization framework based on homomorphic filtering and reflectance ratio in DWT domain for face recognition under varying illumination conditions. Homomorphic filtering (HF) is applied for reducing illumination effect along with contrast enhancement and intensity range compression in face images. Then, illumination deviations are annulled by using reflectance ratio (RR), which yields appropriate texture smoothing and edge preservation. We denoted HF and RR based illumination normalization framework by HFRIN. Further, selective feature extraction by DWT (SFDWT) is performed on HFRIN based face images that discards noise effect. It outcomes in illumination normalized significant facial features, on which subspace analysis is performed to generate small size feature vectors for classification. Various Experiments on benchmark databases such as CMU-PIE, Yale B and Extended Yale B database are performed to assess the robustness of proposed face recognition technique under diverse illuminations.
Why is it important?
The current scenario of face recognition under different indoor and outdoor lighting variations is still facing issues in achieving high performance on large scale database. Many empirical studies highlighted that changes in appearance of a person due to illumination deviations are much more as compared to the face image of another person. Due to varying illumination conditions, variances in the face images are irregular and cannot be modeled precisely. More so, there are infinite degrees of freedom with change in illumination conditions. Thus, a large number of face images are required to be produced per subject to handle such variability in illuminations. However, under extreme lightings conditions, obtaining a large amount of training data increases the computational efficiency. This makes illumination invariant face recognition very difficult for real time face recognition. Therefore, the main objective of this work is to design efficient illumination normalization techniques to annul various types of illumination variations in face images. A good illumination normalization method must have capability to preserve edges in low frequency fields and significant information preservation in high frequency fields. Therefore, we have utilised multiscale analysis of low frequency and high frequency coefficients of face images using wavelet transform. In summary, the area of imaging and face recognition is spreading quickly in the last decade, and this kind of work may help to improve the future of many research because the most important limitation faces the researcher in security area is the minimum acceptable error rate in least processing time (real time processing). The work presented here helps a lot to pass the barriers.