| 摘要 |
A method of image recognition consists in that image noises are removed, image is binarized, image objects are selected and edges of all selected image objects are highlighted. Pulse sequences defining each object are formed, first object comprising utmost number of elements forming the object edge is selected. Second object comprising elements forming the edge thereof is selected, number of elements is less than number of elements forming the edge of first selected object and more than number of elements forming other selected image objects, the distance between first and second selected objects is calculated. Third object is selected comprising elements forming the edge thereof which number is less than number of elements forming the edge of second selected object and more than number of elements forming other selected image objects, the distance between first and third selected objects is calculated. In the same manner in sequence all other objects are selected according to number of elements forming the edges thereof, and distances between each of those and first selected image object are calculated. The object, distance from which to first selected image object is lesser, is arranged first, if objects have equal number of elements and equal distances to first selected object, the object which first arranged by direct motion of hour hand of circle with centre on first selected object, pulse sequence is formed comprising groups of pulse sequences, each defining separate selected object. Groups of pulse sequences are arranged in order of image objects selection, time intervals between each group of pulse sequences are specified, those correspond to distance value between first selected object and selected object defined by specified group of pulse sequences. First pulse at each group of pulse sequences is formed, amplitude value of this one corresponds to direction of object arrangement relative to the first selected object, thresholding of formed pulse sequence is performed, groups of pulse sequences which amplitude and duration less than specified threshold values are removed. Time intervals formed before are removed, pulse sequence is transformed to number sequence which is compared with reference sequence, if those coincide the image is considered as recognized one, and if there is no coincidence cyclic shift of sequence is performed and at each shift stroke comparison with reference sequence is performed, and if there is no coincidence new reference standard is formed. |
