Abstract
Face recognition has received enormous fame in the field of pattern recognition and computer vision. Being a demanding area intensive research has been done by many researchers for more than a decade. However no standard technique exists for extracting the significant features of facial images in different categories. Techniques found in literature produces high accuracy but are computationally expensive which are not applicable in real time applications. In this paper, two well known methods, Discrete Wavelet Transform (DWT) and Weber Local Descriptor (WLD) are used to extract the face discriminative features. First for both types of features, the recognition accuracy is separately measured. In the next step, both types of features are fused using the concatenation method to improve the accuracy rate. To select more discriminative features and reduce data dimensions, computationally efficient algorithm (Kruskal-Wallis) is used. In the last step, three classifiers (SVM, KNN and BPNN) ensemble to improve the accuracy rate. Proposed technique is more efficient in terms of time complexity as compared to GA and PSO. Yale face database is used for all experiments. The proposed technique is highly robust to facial variations like occlusion, illumination and expression change and computationally efficient as compared to existing methods.