| 摘要 |
FIELD: information technology. ^ SUBSTANCE: method involves steps on which a picture of the lateral surface of a bullet (or case) is taken. The picture is broken into fragments of given dimensions. A histogram of phase distribution of brightness gradient vectors in all points of the fragment is obtained for each fragment. The phase histogram is transmitted to the input of a pre-trained neural network, through which each fragment is classified as a trace or noise. On picture fragments attributed to traces, the histogram is broken into ranges on brightness. The longest parallel lines are identified in each range. The inclination angle of lines along which there is maximum accumulation of points with the same brightness is determined. In case of inclination angle of 90, the given trace on the fragment is associated with a primary trace. Histograms of distribution of inclination angles are constructed for all fragments. The maximum of histograms which correspond to the defined texture of the traces is determined, and for all traces, except the primary trace, the type of the trace is determined through the topological position of traces relative the position of boundaries of grooves. ^ EFFECT: wide range of recognised features of images of objects, high probability and accuracy of correct recognition and shorter recognition time. ^ 8 dwg |
