Face detection and identification in real-world settings present many challenges due to lighting variations, pose variances, and occlusion of the face. In this paper, we provide a strong system for detecting and identifying sports celebrities using face images with both traditional machine learning and deep learning techniques. The proposed system employs two different face detection methods (Haar Cascade Classifier and Max Margin Object Detection (MMOD)) and two different recognition methods (Local Binary Patterned Histogram (LBPH) and Pruned ResNet) as CNN-based techniques to identify faces [1].

A sports celebrity image dataset has been used for an extensive series of experiments. This dataset has also been used in conjunction with a publicly available dataset containing images from 31 celebrities in frontal views for face detection and recognition. Multiple combinations of methods have been tested and evaluated to determine the most appropriate model for the recognition of faces in images. The analysis of the results shows that the MMOD detector using a Pruned ResNet model provides the best results in terms of accuracy [2].

The performance differences of structured datasets contrastingly to unstructured real-world image data is compared in addition to a very important need for strong feature extraction/recognitiondetection techniques. Proposed system is highly successful at increasing the efficiency of recognition in unfavourable situations making it practical for potential applications including; surveillance, sports analytics and automated identity recognition [3].

Keywords - Face Detection, Face Recognition, Multi-task Cascade Object Detection (MMOD), Haar Cascade, Local Binary Pattern Histograms (LBPH), Convolutional Neural Networks (CNN), Pruned ResNet, Sport Celebrities, Image Processing.