A Robust Face Recognition System in Image and Video

Abstract

Face detection and recognition has always been one of the research interests to researchers in the field of the biometric identification of individuals. Problems such as environmental lighting, different skin color, complex background, etc affect on the detection and recognition of individuals. This paper proposes a method to enhance the performance of face detection and recognition systems. Our method, basically consists of two main parts: firstly, we detect faces and then recognize the detected faces. In the detection step, we use the skin color segmentation combined with AdaBoost algorithm, which is fast and also more accurate compared to the other known methods. Also, we use a series of morphological operators to improve the face detection performance. Recognition part consists of three steps: dimension reduction using Principal Component Analysis (PCA), feature selection using Linear Discriminant Analysis (LDA), and k-Nearest Neighbor (K-NN) or Support Vector Machine (SVM) based classification. Combination of PCA and LDA is used for improving the capability of LDA when a few samples of images are available. We test the system on the face databases. Experimental results show that the system is robust enough to detect faces in different lighting conditions, scales, poses, and skin colors from various races. Also, the system is able to recognize face with less misclassification compared to the previous methods.

Comparing Performance of Supervised Learning Classifiers by Tuning the Hyperparameter on Face Recognition

“…Another one is Sharma and Sachdeva [41] who obtained an accuracy value of 98.7% by utilizing SVM as a classifier for face recognition. Prakash and Singh [42] reviewed numerous papers and concluded that SVM has a more effective technique than the others and SVM can be trained successfully for face recognition and SVM is also a better learning algorithm compare to the nearest center approach in terms of face recognition, And the last, P. S. Hiremath et al [43] and A. Tofighi et al [44] have been comparing the SVM and K-Nearest Neighbor (KNN) for face recognition and got the result that SVM performs better than KNN with the scores are 99.2% on 3D face recognition [43] and 93.5% on recognizing the face of an image and a video [44].…”

Image Training, Corner and FAST Features based Algorithm for Face Tracking in Low Resolution Different Background Challenging Video Sequences

“…Both single and multiple faces are detected and the face(s) need not to be part of starting frame/image in the video sequence. Dilation and erosion [2] operations are applied on the detected face(s) to obtain borders. FAST (Features from Accelerated Segment Test) features [3] and Harris corner points [4] are extracted from ROI and borders.…”

“…Wiener filter is one of the most powerful noise removal approach [9], which has been used in several applications. Dilation and erosion are morphological operations [2] that are dual to each other i.e. erosion operation shortens foreground, and expands background; dilation operation expands foreground and shortens background.…”

Detection and Classification of Signage’s from Random Mobile Videos Using Local Binary Patterns

“…SVM is basicaly designed to perform binary classification. For multiple classes SVM is constructed using K-problems like one-against-one, one-against-all, the time compexsity of both approach is same, where the evaluation of K class changes for one-againstone(requires K-1 SVM), and for one-against-all(requires K SVM) [12] [16]. In our expirement we use one-againsall method with liner SVM classifiction.…”