摘要
Ada Boost分类器训练算法对特征搜索的时间复杂度较高,改进的PSO-Ada Boost算法采用最佳特征搜索方式训练耗时减少,但在迭代过程中容易陷入局部最优解。为此,提出用混沌粒子群优化Ada Boost训练算法的CPSO-Ada Boost算法。通过引入混沌优化序列增加种群的多样性并扩大粒子搜索范围,帮助粒子克服"惰性"摆脱局部最优解,从而在训练分类器时可以快速寻找到性能更好的弱分类器。在MIT样本库上训练人脸检测分类器结果表明,CPSO-Ada Boost算法减少了训练过程中所需要的特征数量,缩短了训练时间,有效地提高了人脸检测率。
The time complexity of AdaBoost classifier training algorithm search feature is high, and the PSO-AdaBoost algorithm search for the best features can reduce training time-consuming. However, the iterative process is easy to fall into local optimal solutions. To this end, CPSO-AdaBoost based on chaotic panicle swarm optimization AdaBoost training algorithm is proposed. By introducing the chaos optimization sequence increase the di- versity of population and expand the range of particle search range, it can help panicles to overcome the ＂inertia＂ and get rid of local optimal solu- tions, and it can quickly find the weak classifiers with better performance when training a classifier. On the MIT sample library, training face detection classification results show that CPSO-AdaBoost algorithm can reduce the number of features needed for the training process, reduce the training time, and effectively improve the human face detection rate.
出处
《电视技术》
北大核心
2014年第19期175-178,187,共5页
Video Engineering
基金
北京市教委学术创新团队项目(PHR201106149)
