Modeling Human Recognition of Deformed Maps

In this paper, the authors presented the challenges in modeling human recognition of deformed maps and attempted to model human recognition using deformed maps of prefectures based on the values observed in previous studies as the values that our model should imitate. First, the similarity values between the original and deformed shapes were calculated using several distance functions, and the relationship between the similarity and observed values was visually analyzed. Consequently, the dynamic time warping (DTW) distance was adopted as the distance function. The sigmoid function was used to fit the similarity values to the observed values. However, through trials, the authors concluded that a single fitting function could not adequately represent the human cognitive abilities. Therefore, a model consisting of several submodels was constructed. Each submodel fit a cluster of prefectures depending on their characteristics. To determine the appropriate number of clusters, the authors examined the errors between the model output and the observed values for different numbers of clusters. Consequently, the human deformation map recognition ability was modeled with an average error of approximately 20–26% for nine clusters. For some prefectures, the error can be reduced to approximately 10-15%.