摘要 |
PURPOSE: To provide a shape recognition device which can recognize shape more correctly even if the shape includes a noise of handwriting input, etc. CONSTITUTION: The border line of a handwritten figure drawn with a mouse, etc., is inputted on-line from an input device 11, and a feature extraction part 12 performs sample point extraction processing, gravity center calculation processing, intersection detection part processing, projection line removal processing, convex closure processing, sample point determination processing, and tangent vector determination processing as to the handwritten figure. A fuzzy processing part 13 calculates the internal angle between tangent vectors, classifies it into an acute angle, a right angle, an obtuse angle, a wide obtuse angle, a straight line angle, and a superior angle, and calculates the total of respective membership values. A neural network 14 learns shape samples previously and recognizes a triangle, a quadrangle, and a circle from the total of the membership values as to the input figure. If the figure can not be recognized or misrecognized, relearning is performed on the basis of the figure. A BSW or BP method is used for the (re)learning. |