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

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.