A01Feature representation mapping between deep neural network and psychophysical texture　properties

 In the field of computer vision, deep convolution neural networks (CNN) show good performance for the task in the field such as classification, and segmentation, however, we don’t have clear answers for the questions such as “Which feature is effective for the task?”, and “Does the CNN have the relationship to our visual recognition mechanism?”. In this project, we treat texture classification task and try to discuss the questions by use of comparing the inner representation of the CNN and psychophysical texture properties.