Abstract
Haiku is a short poem, but it can be reminded of the scene. A new entertainment can be invented if a computer can represent the scene. This paper reports on prototype system of composite image generation for Haiku. The system processes it as follows. It extracts feature words from the haiku, and searches for images by those words as query on the Web. Then it elects images that suit feature words by the image analysis, and compounds these images. To achieve the system, the technologies of morphological analysis, extracting feature words, clustering, region segmentation and blending are needed.
