Researches On The Straight Line And Curve Detection In Complex Scenes

Straight line and curve is an important feature of the object to be identified indigital image, and it has very important significance on extracting out of straight lineand curve from image effectively and accurately for accurate reconstruction of objectrecognition model, so the detection of line and curve is an important task of imageprocessing and image analysis.Hough transform is a common tool for detecting and extracting the locationparameter of straight line and curve, which has the advantages of robustness, the partof feature points missing does not affect the extraction of object’s feature. However,using Hough transform in detecting straight lines and curves in complex scenes canappear the spurious peak value, and then detect the false line. And the formation offalse peak is caused by too many texture region points in the binary image.For the false peak problem of Hough transform, this paper presents threedifferent ways of weighting Hough transform in order to achieve the goal ofsuppressing false peaks, significantly improved peak detection rate in the real line.The first aspect starts with basic features of original image, calculated by usingthree different kinds of mathematical method of Harris operator, SUSAN operator andvisual saliency method of frequency-tuned which is based on the image of each pixelin the image forming edge contributions, and the calculated results are put into thevoting process of Hough transform, so that the pixels at the edge of the area for alarger weights and texture as well as background areas get smaller weights so thatultimately achieve the purpose of suppressing false straight; The second aspect startswith the neighborhood of the binary image’s feature points, by observing of theneighborhood of the feature point,if the center falls substantially straight line passingthrough the center, the center of the feature point is more likely through the binaryedge image feature points observed neighborhood part of the line, on the basis of thisidea put forward the concept of online-offline (OOR) ratio of the binary image, and inaccordance with this concept, the Hough transform based on the OOR weighted andLVOOR weighted are proposed, the method also reached a peak significantly improvethe detection rate of the true purpose of a straight line; The third aspect starts with theperspective of the distribution of different regions of the binary image’s feature points,on the basis of random sampling of probabilistic Hough transform feature points, taking the binary image of blocks and sub-blocks of different regions have to take adifferent point sampling strategy, that is randomly distributing less feature points insub-blocks which has many the feature points while the regional distribution whichthe fewer feature points ensure have the certain number of feature points, and improvethe probability on the basis of realization of the Hough transform, the method can alsoachieve the purpose of effectively suppressing spurious peak, and increases theoperation speed of the algorithm,it can be used for real-time detection of a straightline.Finally, in contrast to the above several improvements over the algorithm, thisstudy used the weighted Hough transform based on improved SUSAN operator toextract pointer of pointer instrument in order to achieve automatic pointer instrumentreading.