GB1525856A - Image pattern selection and location method - Google Patents

Abstract

1525856 Recognizing images PHILIPS ELECTRONIC & ASSOCIATED INDUSTRIES Ltd 30 July 1975 [2 Aug 1974] 31895/75 Heading G4R The invention relates to generating density change contours on a two-dimensional picture and determining the position of a best match of these contours with a template contour, e.g. the outline of an object being searched for. An image such as an X-ray image of a heart is scanned using a television camera and the signal produced is sampled to produce a plurality of signals each indicating the optical density of an image point in the image. The signals are stored in a matrix store alongwith co-ordinate values indicating the positions of the signals in the store. A value representing the horizontal rate of change at each point is obtained by adding the density values of several matrix positions on one side of the point and subtracting this sum from the sum of the same number of matrix positions on the other side of the point (Fig. 1, not shown) and storing the result. The results are examined to determine the maxima and if these are above a threshold value they are replaced by a "1" value and the other values are replaced by a zero value to form a contour matrix. The subtraction may produce maxima with positive or negative signs and these signs can be stored with the "1" value as in Fig. 4 where the pattern indicated by squares is the contour from the image being examined and the solid squares indicate maxima with negative signs and the open squares indicate maxima with positive signs. A template 'or reference contour is indicated by the triangles. Separate memories may be used to store the positive and the negative patterns. A distance matrix is produced by generating in a matrix store zero values at all locations corresponding to the contour determined from the radiograph as shown in the heavy line in Fig. 3. The adjacent memory locations are filled with a "1" value and the next adjacent locations with a "2" value and so on. Values in this distance matrix at points on the template or reference contour are then added to produce an out of register measure. The template contour is then effectively shifted and the process repeated to produce an out of register measure for each shift and the comparison with the lowest sum is taken to indicate the best match.