Research On Line Feature Extraction Based On Beamlet Transform

Line feature detection plays a very important role in computer vision, patternrecognition and image processing area. Line features play an important role in all of thefeatures. The traditional line feature detection methods treat image in pointwise way,which disobeys the human vision.In order to effectively represent and deal with some high dimensional data, linefeature extraction based on multiscale analysis have been developed recently. Beamlettransform is a kind of multiscale geometric analysis method. It uses line as basic unitwhen performs integral operation, then we get image line feature. The method conformsto the basic features of the human visual which identify objects. This paper analyzesbasic theory of the Beamlet transform and in-depth analysis Beamlet algorithm. Themain works and achievements are as follows:1. Anew line feature extraction algorithm based on improved Beamlet structurelessalgorithm and Canny operator is proposed. First, Beamlet transform is performed. Thereis at most one optimal Beamlet in a dyadic square by improving Beamlet structurelessalgorithm，and using the new drawing rule and new energy function. Second, use Cannyoperatator detection edge by selecting larger Sigma in order to detect obvious edge.Finally, line feature is detected by combination of both. The evaluation of this algorithmis taken from several aspects, such as the continuity of the line feature extraction, thewrong detection and miss detection. Moreover, compare this method with some existingmethods. The experimental results show that our proposed method not only overcomestheir weakness such as fractured, overlapping, obscure, false edge and so on, but alsoeffectively improves the accuracy and continuity when extracts line feature of compleximage.2. A new pavement crack detection based on improved Beamlet tree-structurealgorithm is proposed. In order to overcome the large computation of Beamlettree-structure algorithm, we propose an improved algorithm. And apply it to pavementcrack algorithm, therefore, it solve the problem of pavement crack detection which hasthe disadvantage of bad noise immunity and inaccurate test results. First, the image ispreprocessed to correct the nonuniform background illumination by calculating themultiplicative factors that eliminate the background lighting variation. Then, the imageis transferred to the binary image using Otsu’s thresholding segmentation algorithm. At last, extract pavement crack from the binary image using the tree-structure of Beamlet’smutiscale, which transform a “bottom to top” inter-scale inhibition to optimize theobject function to a “top to bottom”. Therefore, the improved method reducescalculation complexity and time. Experimental results show that the proposed methodcan extract crack form the complex pavement background and noises quickly, moreover,it keeps the continuity of the crack as well.