中文摘要：

直线段特征是数字图像中的重要信息，具有丰富的语义．直线段提取方法的优劣直接影响到高层次图像处理的效果．针对当前直线段提取方法中存在的直线断裂问题，改进现有算法，提出一种基于假设验证的稳健的、高准确度的直线编组算法．该算法从几何相似和纹理相似性两个方面对直线段进行约束，如果满足条件，则假设分裂的直线段属于同一条直线，而后在断裂区域进行二次边缘直线提取，并验证所提取的新直线与原有断裂直线段是否属于同一直线．本算法在一定程度上能够避免直线编组中通常存在的错连与漏连问题，通过对航空影像的试验验证了该方法的有效性．

英文摘要：

Line segment feature is a very important information in digital image because it contains abundant semantic information. The positive and negative of the line segment extraction method directly affects the processing effects of high-level image processing. The common problem of the current line extraction methods is that the extracted line segment were always discontinuous. In order to resolve this problem, a robust approach for detecting straight line segment based on perceptual organization is presented. This algorithm combines two important steps： hypothesis and verification. At first, in hypothesis step, the fractured line segments were organized into a line according to geometrical constraints and textural features constraints, and then were supposed to belong to a line if they satisfy the constraints. In verification step, edge were detected in region of fracture again and the extracted new line were tested to belong to the former line or not, if the new line belongs to the former line, the former line and the new line are considered as one line. The flow of this paper is as fellows. Firstly, geometrical constraints and textural features constraints were set to constraint which lines may belong to same line. Geometry constraints include syntropy distance between two lines, angle between two lines, side distance between two lines and overlapping degree between two lines. Secondly, the lines that meet geometrical constraints and textural features constraints were considered as one line. Thirldly, the lines were verified to belong to the same line or not. At last, experiment was done to aerial images that contain many line features, and the experiment results show that the proposed algorithm is more efficient, which can also solve the straight segment fracture existing in the current methods.