In this paper, we describe a new image retrieval system for scenery images named IRIS (Image Retrieval by Impression words and Specific object names) which uses specific object names in addition to impression words which can reflect ambiguous human kansei (impression and sensitivity) as the retrieval keywords. IRIS has not only the function which retrieves images but also that can estimate keywords to images automatically. In IRIS, first an image is divided into some regions. Next each region is roughly classified into the sky, earth and water categories by the image recognition method using a neural network. Then the image characteristics are extracted from each category, and the impression words are given to the image automatically. After the regions are classified into sky/earth/water categories, they are classified into much more detailed objects such as mountain or cloudy weather sky. By the hierarchical recognition, the specific object names are given to an image automatically. Using concrete object names with vague impression words as keywords, flexible retrieval is possible. We carried out extensive experiments to test the performance of the developed IRIS. The retrieval experiments has shown the efficiency of retrieval when the impression words and the specific object names are used together. Also, they have shown the high retrieval performances of IRIS.
ASJC Scopus subject areas
- Computer Science Applications
- Cognitive Neuroscience
- Artificial Intelligence