Research On A Multiscale Line Segment Detection Algorithm

Line segment is a common geometric feature in real scene,which is widely used in var-ious visual tasks.The existing mainstream algorithms detect line segments in a single scale.Although the detection algorithm chooses a low scale value to ensure the accurate position of the line segment,it will produce a large number of redundant detection line segments.And the algorithm chooses a high scale can effectively suppress redundant information,but the position of the line segments will shift.Therefor,we can fuse the features in multiple scales to improve the line segment detections.An investigation of the above-mentioned topics is conducted in this paper.The main contributions and innovations of our work contain the following three parts.According to the characteristics of line segments in different scales,a multi-scale line segment detection algorithm is proposed.First,the scale space representation of the image is obtained by gaussian filtering,and the gradient information are calculated.Then,region grow is carried out under each scale to get the line-support regions.A robust region is ob-tained by fusion processing to reduce error results.Experimental results show the proposed algorithm has better performance subjectively and objectively than the existing line segment detection algorithms.A new method of line segment verification is proposed.The interference information in the image will lead to the error detection of line segments.The current line segment verification method is based on its own reliability,and ignores the relationship between the line segments.Therefore,we propose a global verification method based on the hypothesis of quasi-manhattan world,which takes the correlation between line segments as the basis of reliability analysis.Experimental results show that,compared with the existing verifica-tion methods,the new verification method can filter the error detection line segments more effectively,and significantly improving the accuracy of results.A reasonable evaluation system is established.In the past research and development,there is a lack of an objective and reasonable evaluation system.We evaluate the line seg-ments from two aspects:subjective evaluation and objective evaluation.In the subjective evaluation,we gives a systematic definition of the detection effect of line segment.In the objective evaluation,the objective evaluation model is improved,and the evaluation algo-rithm is designed and implemented.The effectiveness of the proposed objective evaluation algorithm is verified by experimental analysis.Finally,we give the evaluation results of the line segment detection algorithm and the comparison algorithms.