Study on Controlling Emotion of a Multimedia Viewer Based on Interactive Evolutionary Computation and Physiological Analysis

我们在2005-2007财政年度分三步进行了这项研究。在第一步“提取可能影响生理反应的电影特征”中，我们首先从几种电影中剪切了60个20秒的场景，并提取了78个物理电影特征。其次，我们构建了一个由（紧张-放松）轴和（振奋-沮丧）轴组成的二维心理空间，称为情感空间，并从78个可能的电影特征中筛选出30个，而不是使用生理学实验，而是使用心理学实验。在第二步“提取电影特征与生理反应之间的关系”中，我们首先用30个特征中相关性最高的一个特征的20个不同变量来制作抽象的实验电影场景。由于电影语境即故事对人的情感有着深刻的影响，我们制作了抽象的电影场景来避免它的影响。其次，我们通过展示抽象影像来测量生理数据。测量数据为两侧和背面两个点的脑电图、心电图、出汗率、血压水平、手指脉搏量。然而，被试之间测量数据的差异大于物理电影特征值的差异，我们还没有发现电影特征与生理反应之间的显著关系。在第三步中，“使用扩展交互进化计算（IEC）建立驱动情感”，我们的目标是建立一种优化电影特征的技术，以驱动电影观众的生理状态到给定的目标状态。由于在第二步中我们还没有发现驱动人体生理的电影特征，我们将重点放在改进IEC技术上。我们所做的是将IEC与进化多目标优化相结合，引入粒子群优化代替进化计算，引入IEC用户评价模型等。

We conducted this research with three steps in 2005-2007 fiscal years.In the first step, “extracting possible movie features that may influence physiological responses, “we firstly cut 60 scenes of 20 seconds from several varieties of movies and extracted 78 physical movie features. Secondly, we constructed a 2-D psychological space, called an emotional space, consisting of(tension-relax)axis and(refreshing-gloomy)axis, and narrowed 30 from 78 possible movie features not using a physiological experiment but a psychological one.In the second step, “extracting relationship between movie features and physiological responses, “we firstly made abstract experimental movie scenes with 20 different variables of one feature that shown the highest correlation among 30 features. Since movie context, i.e. story, deeply influences to human emotion, we made the abstract movie scenes to avoid its effect. Secondly, we measured physiological data by displaying the abstract movies. Measured data are EEG at two side points and two back side, ECG, sweating rate, blood pressure level, and finger pulse volume. However, the difference of measured data among subjects was bigger than the difference of values of physical movie features, we have not found out the significant relationship between the movie features and physiological responses yet.In the third step," establishing driving emotion using extended interactive evolutionary computation(IEC),"we aimed to establish a technique to optimize movie features to drive a movie viewer's physiological condition to a given target condition. Since we have not found out the movie features that surely drive human physiology in the second step, we focused on improving IEC technique. What we did was combining IEC with evolutionary multi-objective optimization, introducing Particle Swarm Optimization instead of evolutionary computation, introducing IEC user's evaluation model, and others.