| 摘要 |
PURPOSE:To improve the accuracy of image identification by covering the whole multidimensional image with multidimensional polyhedra which overlap with one another, determining a weight function which is large in the center of each polyhedron and decreases toward the peripheral part, and finding the weighted sum of respective picture elements from the weighting function and obtaining the mask output of the image. CONSTITUTION:The whole image is covered with rectangular areas which overlap with one another. For example, an area of size aXb at the left upper corner is covered with small areas 7, 8, 9, and 10 of size 2aX2b. The value of a picture element having coordinates (x, y) is denoted as P(x, y). The weight function Wij(x, y) is determined for the small area 7, etc. The function Wij is maximum in the center of the small area and decreases toward the peripheral part to reach 0 outside the small area. Here, an output L(Xi, Yj) corresponding to the coordinates (xi, Yj) is expressed by L=SIGMAWijP(x, y). Then the sum covers all the picture elements. A template matching method is applied to this output L. |
