READING METHOD FOR CHARACTER IN IMAGE AND RECORDING MEDIUM WHERE SAME METHOD IS RECORDED

摘要

PROBLEM TO BE SOLVED: To read (recognize) characters in an image even at a specific line of sight by calculating correspondence points between videos for a photographed image from distribution features of colors, and generating a multiple-composite image by putting images one over another. SOLUTION: When a signboard image is present in a scene image photographed by a camera 101, its image data is inputted. An area which includes characters is extracted from the inputted image data. Distortion due to a three-dimensional tilt of an image of the extracted character area is corrected. The tilt-corrected image data are stored in a data storage part 103B and when there is no input image to be processed next, correspondence points between videos are calculated for the tilt-corrected image data from distribution features of colors; and images are put one over another according to the calculation result to generate a multiple-composite image. Character patterns are recognized from the multiple-composite image and the recognition results are stored in a data storage part 103B.