tailieunhanh - Fast gaussian distribution based adaboost algorithm for face detection

This paper inherits face detection framework of Viola-Jones and introduces two key contributions. First, the modification is used to apply integral image so that features are more informative and help increase detection performance. The second contribution is the new approach to utilize AdaBoost that uses Gaussian Probability Distribution to compute how close to the mean positive and negative distributions are, then classify them more efficiently. | ISSN 1859-1531 - THE UNIVERSITY OF DANANG, JOURNAL OF SCIENCE AND TECHNOLOGY, NO. 6(127).2018 45 FAST GAUSSIAN DISTRIBUTION BASED ADABOOST ALGORITHM FOR FACE DETECTION Tuan M. Pham1, Hao P. Do2, Danh C. Doan2, Hoang V. Nguyen2 University of Science and Technology - The University of Danang; pmtuan@ 2 Hippo Tech Vietnam; {haodophuc, , }@ 1 Abstract - In the past few years, Paul Viola and Michael J. Jones have successfully developed a new face detection approach which has been widely applied to many detection systems. Even though the efficiency and robustness are proved in both performance and accuracy, there is still a number of improvements that we can apply to enhance their algorithm. This paper inherits face detection framework of Viola-Jones and introduces two key contributions. First, the modification is used to apply integral image so that features are more informative and help increase detection performance. The second contribution is the new approach to utilize AdaBoost that uses Gaussian Probability Distribution to compute how close to the mean positive and negative distributions are, then classify them more efficiently. Furthermore, by experiments, we also prove that a small fraction of a feature set is far enough to develop a good strong classifier instead of the whole feature set. As a result, the memory required as well as the time for training is minimized. Key words - face detection; Gaussian distribution; AdaBoost; Haar-like pattern; weak classifier 1. Introduction In recent decades, along with the rapidly advanced improvement in technology, face detection has now become the most popular topic that can be applied to many fields in industries or in real life. Algorithms for face detections are developed quickly and become more enhanced to support complicated applications like multiview face detection [1-4], occluded face detection [3, 5], pedestrian detection [6, 7], . In this paper, we inherit .