Lines Detection And Reconstruction Based On Multi-scale Analysis

The lines can find in many kinds of images, just as blood vessels in medical image, rivers and roads in aerial imagery, characters and words in text image and structure light lines3D measurement. Lots of algorithm studies have been made on them. The detection and reconstruction of lines feature plays a decisive role in low-level computer vision, whereas the accurate detection and reconstruction lays the foundation for image understanding and3D’s reconstruction in high-level computer vision.This paper describes the multi-scale space theory and the application in the image processing. At present, the rapid development of the algorithm of image processing based on multi-scale space theory, the core idea of which lies in the understanding of different goals resulting from different resolutions. The target of strong structures and weak ones is analyzed in low-level resolution and high-level resolution respectively. Therefore, the main advantages of the method based on multi-scale space theory are high speed processing and high increased robustness.Aiming to lines reconstruction and understanding in images. Laplacian of Gaussian with multi-scale and multi-angle adaptive strengths the lines based on multi-scale space theory, obtains lines scale, angle information and achieves centerline detection using non-maximum suppression and double threshold, tracking and thinning, break connection and junctions connection after centerline detection. Using scale and angle information, the lines is reconstructed at last. The experiment results indicate the validity of lines reconstruction algorithm.There are many important and difficult problems in the processing of lines detection and reconstruction, and the typical one is lines junctions. This paper is dedicated to make a study on lines junctions with two branches, such as "X-junctions","T-junctions","V-junctions" and "L-junctions". In order to distinguish junctions and lines, the improvement of original square filter is employed based on receptive field attention mechanism, though the differences of junctions center point filter response. Research on junction has been labored to probe that they are effective supplements to the algorithm of lines detection and reconstruction.