PURPOSE: To achieve an accurate detection without being affected by noise which is contained in an image using a neural network for taking in the property of the target image by learning.
CONSTITUTION: A signal indicating the shade level of nine elements inside a minute region A is read from an image storage 18 and is input to a neural network 22 and then a signal indicating the direction of a contour line in the region A is output from an output layer 34. Then, when the scanning of the region A is completed, the possibility of presence of the contour line of each element is evaluated, and the contour line is synthesized and stored. Also, a certain addition value corresponding to a position is accumulated as an evaluation value indicating the possibility of presence of the contour line by an evaluation value addition means 24 and a contour line criterion means 26 judges for each picture element whether the contour line exists or not by a contour line criterion means 26. Further, a high detection accuracy can be obtained by providing a plurality of output nerve cell elements to the neural network 22.
YOSHIZAKI TETSUYA
UCHIKAWA YOSHIKI
YOSHIZAKI TETSUYA
UCHIKAWA YOSHIKI