WSEAS Transactions on Signal Processing
Print ISSN: 1790-5052, E-ISSN: 2224-3488
Volume 10, 2014
Towards Enhancing the Face Detectors Based on Measuring the Effectiveness of Haar Features and Threshold Methods
Authors: , ,
Abstract: Face detection has been regarded as the most complex and challenging problem in the field of computer vision, due to the large intra-class variations caused by the changes in facial appearance, lighting, and expression. Face detection is the essential first step towards many advanced computer vision, biometrics recognition and multimedia applications, such as face tracking, face recognition, and video surveillance. One of the most famous approaches that is successful is the Viola & Jones algorithm. In this paper, systems were designed based on this approach to measure the effectiveness of the different Haar feature types, and to compare two types of threshold computing methods. The two methods used for computing thresholds are the average of means and the optimal threshold methods. There are 8 different Haar features has been used in building these systems. The implemented systems have been trained using a handpicked database. The database contains 350 face and nonface images. Adaboost algorithm has been used to build our detectors. Each detector consists of 3 cascade stages. In each stage, we randomly use a number of weak classifiers to build the strong classifier. Each weak classifier is computed based on threshold before entering the Adaboost algorithm. If the image can pass through all stages of the detector, then the face will be detected. The detectors have been tested using the CMU+MIT database. Some recommendations have been suggested according to the Haar features and the computed threshold to improve the face detection of Viola Jones approach.
Search Articles
Keywords: Face Detection, Haar-Like Features, Pattern Recognition, Weak Classifier, Integral Image, Strong Classifier, Adaboost Algorithm.
Pages: 662-673
WSEAS Transactions on Signal Processing, ISSN / E-ISSN: 1790-5052 / 2224-3488, Volume 10, 2014, Art. #68