影响触觉图像识别因素的量化分析

Multi-factor Analysis Assisting T-Image Design for Tactile Cognition

为了使更多盲人能受益于盲文书籍所伴随的插图,区别于传统的V图像(视觉图像),对设计适合触觉认知的T图像(触觉图像)提出新的设计原则.首先将242张常见物品的V图像制作为线条凸起的可触摸图片;然后邀请10位盲人被试和10位蒙眼明眼人被试通过触摸来尽量准确地命名这些线条图,并要求被试在触摸的过程中进行"出声思维";再根据被试对线条图的描述,提取22个可能影响二维线条图触觉识别的特征;最后以识别正确率作为图片识别难易程度的指标,使用随机森林算法进行了特征建模,并对所有特征进行单因素和多因素的回归分析.实验结果表明,通过随机森林算法建立的模型,可以基于图片中这些特征预测图片触觉识别的难易程度;通过多因素回归分析,提取出对触觉识别有显著影响力的几个重要特征,并用于指导T图像的设计.

In order to help more blind people to benefit from accompanying illustrations while reading, westudy how to design a T-image （tactile image） suitable for tactile cognition, which is different from the traditionalV-image （visual image）. Firstly, 242 V images of common objects were made as raise-line tactile images;then, 10 blind subjects and 10 blindfolded sighted subjects were asked to name these raise-line imagesby touching as accurately as possible and to make “Thinking Aloud” during the touching process; after that,according to the subjects’ description, 22 features were extracted which may affect the difficulty of tactilerecognition of raise-line images; finally, we used the random forest algorithm to build a machine learningmodel of the 22 features with the naming accuracy as the index of the tactile recognition difficulty and Single-factor and multi-factor regression analysis were used to compare the importance of the features. The resultsshow that the model based random forest algorithm can be used to predict the difficulty of tactile recognition based on these features of a raise-line image; The multi-factor regression analysis find some featuresthat have significant influence on the tactile recognition, which can guide the T-image design.