Abstract: Components-based cartoon face generation is composed of components combination and features adjustment. These two stages can be solved through combination optimization and continuous optimization respectively. However, when cartoon face feature parameters are optimized, it is difficult to use functions to explicitly express the optimization objectives, which is a typical problem of implicit objective optimization. To solve this problem, this paper proposes the interactive differential evolution algorithm based on opposition-based learning strategy (IDE-OBL), which transforms the interaction method of artificially providing adaptive values by human beings in conventional interactive evolutionary algorithm to the method of pairwise comparison. In this new evolutionary algorithm, opposition-based learning strategies are applied to accelerate algorithm convergence, which will reduce the user evaluation times to some extent. The experimental results show that in cartoon face generation IDE-OBL is better than conventional IGA and IDE which do not use OBL, IDE-OBL reduces evolution iteration numbers and is beneficial for the ease of user fatigue.

Key words: interactive evolutionary algorithm, differential evolution, cartoon face, opposition-based learning

摘要： 基于组件的卡通人脸生成分为构件的组合及特征调整两阶段完成，可分别视为组合优化和连续优化问题解决。然而，人脸特征参数优化过程中很难用函数显性表示其优化目标，是典型的隐性目标优化问题。针对此问题，提出基于反向学习策略的交互式差分演化算法（interactive differential evolution algorithm based on opposition-based learning strategy，IDE-OBL），将传统交互式演化算法中人为提供适应值的交互方式转化为成对比较的方式，采用反向学习策略加快算法收敛，在一定程度上减少了用户评价次数。实验结果表明，在基于组件的卡通人脸生成问题中，IDE-OBL比未使用反向学习策略的IGA和IDE要好，减少了演化迭代次数，有利于用户疲劳程度的缓解。

关键词: 交互式演化算法, 差分演化, 卡通人脸, 反向学习