公开(公告)号：US20040008890A1

公开(公告)日：2004-01-15

申请号：US10193544

申请日：2002-07-10

Applicant: Northrop Grumman Corporation

Inventor： Walter D. Clark ,  Richard L. Hoffman

IPC: G06K009/48

CPC classification number: G06K9/48 ,  G06T7/12 ,  G06T7/181 ,  G06T2207/10016 ,  G06T2207/10048 ,  G06T2207/20132 ,  G06T2207/30212

Abstract: The present invention includes a method for analyzing an image to screen for target candidates wherein an edge-follower detects straight segments within a plurality of longer edges and detects also from the longer edges, contours that go around areas that are the right size. The invention also contemplates starting the edge-follower from seeds on the image and terminating the edge-following process if the edge meanders too much to ever become an acceptable contour. In accordance with another embodiment, the straight segments can be further screened by measuring the gradient direction and then noting the deviation from the mean of those direction values. The highly straight edges thus found, in confluence with each other, or near a correct size contour define likely candidates.

Abstract translation: 本发明包括一种用于分析图像以筛选目标候选的方法，其中边缘跟随器检测多个较长边缘内的直线段并且还从较长边缘检测围绕正确尺寸的区域的轮廓。 本发明还考虑从图像上的种子启动边缘跟随器，并且如果边缘曲折太多以致永远不会成为可接受的轮廓，则终止边缘跟随过程。 根据另一个实施例，可以通过测量梯度方向然后注意到与这些方向值的平均值的偏差来进一步筛选直线段。 因此，如此发现的高度直线的边缘彼此汇合或者靠近正确的大小轮廓确定了可能的候选者。