Image Retrieval based on Object Extraction and Kansei Estimation

摘要

This research proposes an image retrieval system based on object extraction and Kansei (that is, a high-order function of the brain, including inspiration, intuition, pleasure and pain, taste, curiosity, aesthetics, emotion, sensitivity, attachment and creativity) estimation, which enables retrieving images by using object names and/or human Kansei as retrieval keywords. The proposed system focuses on the extracting objects from image and the relationships between objects and human Kansei. It depicts more clearly how human impression causes from characteristics of the extracted objects in images, whereas other methods evaluate impressions from characteristics of the whole or parts of images. Some experimental results in object recognition, human Kansei estimation and image retrieval are presented.