Image corner detection plays an important role in image analysis and recognition. This paper presents a novel corner detector based on the growing neural gas (GNG) network and this proposed detector is called GNG-C. With the GNG network,image topology information can be learned and used to implement corner detection. The GNG-C approach can be described as consisting of the following steps. First,a canny edge detector is used to acquire the contour information of the input image. This edge information is used to train a modified GNG network. A special stopping criterion is defined to terminate network learning. Second,vectors formed between network nodes and their neighbors are used to measure curvatures. Third,dynamic regions of support (ROS) are determined based on these curvatures. These ROS are used to suppress curvature noise. The curvature values of the nodes are then analyzed to estimate the candidate corners. Finally,the candidates are distilled by a non-maxima suppression process to obtain the final set of corners. Experiments on both artificial and real images show that the proposed corner detection method is feasible and effective.
SUN Wei,YANG Xuan College of Computer Science and Software Engineering,Shenzhen University,Shenzhen 518060,China
为了解决面向大规模数量点集时的RPM(robust point matching)收敛时间较长的问题,分析了RPM执行过程中各关键步骤的时间复杂度,针对算法中的矩阵求逆与矩阵相乘进行了基于OpenMP的并行实现;同时针对RPM算法中的运算关系分析了并行实现的可行性,得出它并不适合采用多线程并发以提高算法效率的结论。文中比对了MPI与OpenMP的并行实现效率,并详细分析了高速缓存干扰现象。实验结果表明,该方法可以快速实现点集的匹配,有效地提高了该算法的运行效率。
