High-speed face detection using fuzzy membership function based adaboost algorithm

Face detection is the very well-known problem in Computer Vision applied in any detection and recognition frameworks. With the rapid change in Machine Learning and Deep Learning, this line of research has achieved many profound success that solves the original problem and associated aspects. In this paper, authors revisit the problem in a new approach that enhances the performance and accuracy of the state-of-the-art algorithm developed by Paul Viola and Michael J. Jones. Fuzzy membership functions are utilized in the classification process which helps reduce the amount of calculations as well as the training and classification time. By well conducting experiments, the authors have clearly and convincingly proved the advance of this research compared to the original work. Besides, the performance of each Fuzzy membership function is carefully analyzed to give the insight of the proposed method. In addition, Haar-like patterns inherited from Viola-Jones systems are also extracted in the better way resulting in more informative features. Furthermore, from experiments, authors proved that a small fraction of the whole feature set is far sufficient to build a strong classifier. It leads to the decrease in training time and memory storage is not compulsory as in the previous work. In summary, the proposed method has improved the original solution of Viola and Jones in both accuracy and robustness and also in the training process.