摘要
现有DCT域人脸认证方案未能有效解决掩膜窗口的自适应优化和分量的组合选择;对遗传算法种群中的50个个体进行二进制编码控制人脸图像DCT域分量选择;每个个体初始化选定100个DCT分量,以交叉错误率为适应度函数,通过锦标赛选择、0.8概率的单点交叉和0.02概率的变异操作,全局搜索出具有整体高鉴别性的分量组合;实验结果表明遗传算法优选的DCT分量不仅改进了认证性能,并且对不同数据库具有自适应调节能力。
Neither the adaptive optimization of pre- masking window nor the coefficient combination selection has been effectively solved by the existing face verification schemes in DCT domain. Genetic algorithm was employed to optimize the DCT coefficient selection of face im- age, in which 50 individuals were encoded as binary codes. Tournament selection, one-point crossover with the probability of 0.8 and muta- tion with the probability of 0.02 were performed to minimize crossover error rate, which was considered as the fitness function. The combi nation of DCT coefficient with high holistic discriminant ability was adaptively determined by the global search of genetic algorithm. The ex- perimental results show that the verification performance of DCT coefficients is improved by the selection of genetic algorithm. In addition, the proposed face verification algorithm is adaptive to different databases.
出处
《计算机测量与控制》
北大核心
2013年第7期1983-1986,共4页
Computer Measurement &Control
基金
国家自然科学基金(61262019
61202112
61070163)
中国博士后科学基金(2013M531554
20110491736)
山东省优秀中青年科学家科研奖励基金(BS2011DX034)
贵州省科技厅基金资助项目(黔科合机计字[2013]4002)
关键词
离散余弦变换
遗传算法
自适应优选
分量选择
人脸认证
discrete cosine transform
genetic algorithm
adaptively optimized selection
coefficient selection
face verification
