【作者】 寇强强；

【导师】 许曰滨； 赵志刚；

【作者基本信息】 青岛大学 ， 计算机软件与理论， 2004， 硕士

【摘要】 人脸识别是一项极富挑战性的课题。传统方法需要极其繁多的前期工作和识别阶段的诸多限制，因而不论从理论上还是在实际应用中都极为困难。本文把遗传算法应用于人脸识别的分割、定位和角度矫正三个环节，并建立了相应的数学模型。 具体做法是用二进制数串作为染色体，表示某个分割阈值。遗传算法中的选择算子采用“精英选择”策略，即每一代种群都将上一代中最优染色体保留下来；杂交阶段，从父代和子代染色体中选取最优的两条染色体作为杂交后代；变异算子与传统的取反变异不同，需两条染色体参与，通过逻辑运算，使得种群中同一基因位上的基因不会出现全0或全1的情况，从而最大限度地避免了早熟收敛。本文还添加了一种“倒位算子”，即在一条染色体上随机选择两点构成一个子串，首尾倒置形成新子串替代原子串在个体中的位置。 本文对于提出的遗传算法，给出了遗传算子的数学描述，建立了精确的马尔可夫链模型，并在此基础上给出了遗传算法全局收敛性的证明。 通过仿真实验表明，本文提出的遗传算法与标准遗传算法相比，收敛速度和得到的最优值都有了极大的提高，使人脸识别各个环节避免了繁重的工作，提高了识别的质量和速度。更多还原

【Abstract】 Human face recognition is a challenging issue. The traditional methods of face recognition need most heavy prepare works and many confinements in the real face recognition phase. So in this article, GA is applied in the principal steps of face recognition, including segmenting face, location of face and curing of face angle, simultaneously, the mathematical model is built.The concrete means is to design chromosome by a binary chain comprised by 8 genes, which represent a segmenting threshold that is corresponding to a gray. Selection operator is designed to be ’elite selection’; in crossover phase, ’preferred selection’ is presented; differed from traditional methods, in the course of mutation, at least both a ’0’ and a T exist simultaneously on the same gene position in the current population and that avoid premature mostly. Inversion operator is presented to generate a new binary chain by randomly selecting two gene positions and inversing the chain between the two positions in a chromosome.According to the genetic algorithm presented in this article, amathematical description is presented and an accurate markov model is built. Based on this, the presented algorithm in this article is proved to be global convergent.The emulation illustrates that alongside the standard genetic algorithm, the convergence velocity and optimal solution of the presented genetic algorithm is enhanced mostly. Heavy works in steps of face recognition is so reduced that the effect and rate of face recognition is improved mostly.更多还原