The conventional face recognition systems were designed by considering the facial datasets prepared under the constrained or controlled environment, however practically such methods does not works as the captured face images may suffer from the various challenges such as pose variations, plastic surgery, expressions variations, low-resolution, illumination variations, occlusions etc. The methods proposed for face recognition in recent past failed to address all such challenges for the robust face recognition as the face descriptors designed are mainly either based on the histogram representations or spatial information or both. However the current techniques are inefficient while dealing with unconstrained and plastic surgery datasets. In this paper, we proposed novel hybrid face descriptor that bridges the gap between histogram representations and spatial information efficiently. After the pre-processing, we applied multi-directional dual cross pattern and local directional pattern independently. The outcomes of both techniques fused that generate the unique face descriptor called hybrid face descriptor. Hence hybrid descriptor address the challenges related to variations in pose, expression, illuminations effectively. After the face descriptor, histogram features at different levels extracted. To improve the face recognition accuracy, in this paper we proposed the correlation weights based features selection algorithm which not only performs the features reduction, but also preserves the most unique features for the classification. The experimental result shows that proposed face recognition framework improves the overall accuracy different datasets.
Volume 11 | 01-Special Issue
Pages: 720-729