In the first level process,based on traditional chaincode tracing algorithm,processes such as selection of staring point,chain-code tracing,line segment extraction and so on,are improved to accomplish line segment extraction on the sub-block.

Consequently in higher level,registration of four sets of non-overlapping windows covering the image in on overlapping manner is performed for high efficiency line segment extraction involving only linking and combination of the line segments in sub-block of lower level.

Hough algorithm are more specially dedicated to the long and straight line segment extraction,and are not sensitive to irregular line. Based on Hough algorithm,we apply mathematical morphology operator,which can locate irregular line segment accurately.

The line segment extraction is performed using a standard algorithm.

For line segment extraction, a three-stage algorithm has been implemented.

As part of middle level processing in computer vision, line segments extraction is a bridge between low level processing and high level reasoning and high level computations. Automatically extraction of line segments in a complex scene is very difficult. A new technique that extracts line segments from directed edge images by labeling and growing was proposed. This technique labels on edge pixels of similar directions, and seeks a line segment candidate...

As part of middle level processing in computer vision, line segments extraction is a bridge between low level processing and high level reasoning and high level computations. Automatically extraction of line segments in a complex scene is very difficult. A new technique that extracts line segments from directed edge images by labeling and growing was proposed. This technique labels on edge pixels of similar directions, and seeks a line segment candidate for the current edge pixel in its directed neighborhood. The experimental results show the validity of this technique in the cases of complex scenes. Some defects of this technique and further work directions were also presented.

The paper presents a general paradigm of semiautomatic building extraction from aerial stereo image pair.In the semiautomatic extraction system,the building model is defined by selected roof type input through human_machine interface and input points of rough position of the building by operator.With the geometrical constraints of the model,low_level and mid_level processing including edge detection,straight line segments extraction and line segments grouping are to establish the...

The paper presents a general paradigm of semiautomatic building extraction from aerial stereo image pair.In the semiautomatic extraction system,the building model is defined by selected roof type input through human_machine interface and input points of rough position of the building by operator.With the geometrical constraints of the model,low_level and mid_level processing including edge detection,straight line segments extraction and line segments grouping are to establish the initial geometrical model of the roof_top.However,the initial geometrical model is not so accurate in geometry.To attain accurate results,straight_line extraction and positioning by least squares matching with object_space based geometrical constraints rectifies the initial geometrical model.The adjustment model integrates the straight edge pattern and 3D constraints together and is a well_studied optimal method.The edge model is based on strict zero_crossing definition,and the strict geometrical constraints of the corners of the extracted building as 'knowledge' that is clearly described by the adjustment model are fused with image feature,such as edge and straight line.The model uses all of the information and constraints of the building borders to attain an optimal result,so this is a model driven strategy that matches the roof model with the image pair depending on the criteria of least squares.In principle,the method does not need feature detection,so there are no threshold problemes.After gaining proper initial values,the adjustment model can flexibly process extraction of kinds of roof types by changing or assembling the geometrical constraints in object_space. On the other hand,the paper also discusses the scheme of practical system of semiautomatic extraction.To build a practical semiautomatic extraction system,correctness,accuracy,robustness,speed and interactivity,all these factors should be taken into account.In the paper,a robust algorithm of extracting main direction of the house is addressed.By detecting edge,tracing the edge chain vectors and extracting straight_line segments,a Weighted Histogram of Angles (WHoA) is generated by the angles of extracted line segments.With the histogram,the main direction of the building can be easily detected by searching the max value of the histogram.Histogram method is a statistical manner like Hough transformation; it can effectively remove the influences of gaps between collinear segments and racket segments.The experiments show that that main direction and outer border detection by the WHoA method is robust.The line segments extraction and confirming the main direction of the house could save the operator's input action and offer good initial values of the corners for the consequent adjustment evaluation.

Perceptual organization is a powerful tool for automatic detection objects. In this paper, a perceptual organization approach based on Bayesian Networks is proposed, and used as a method of detection 3-D man made structure objects, such as buildings et al., from urban aerial image. Objects are detected by means of hypotheses verification, and the approach is divided into four steps: edges and lines segments extraction, parallelograms creation, hypotheses creation, and hypotheses verification. Two Bayesian...

Perceptual organization is a powerful tool for automatic detection objects. In this paper, a perceptual organization approach based on Bayesian Networks is proposed, and used as a method of detection 3-D man made structure objects, such as buildings et al., from urban aerial image. Objects are detected by means of hypotheses verification, and the approach is divided into four steps: edges and lines segments extraction, parallelograms creation, hypotheses creation, and hypotheses verification. Two Bayesian Networks are designed for information integration in the process of hypotheses creation, and hypotheses verification respectively. With the learning function from Bayesian Networks, the approach is provided with self-adaptation to some extent. The validity of this approach is indicated by experiment.