基于FPGA的图像多尺度特征点提取及匹配

Multi-scale feature points extraction and matching of images based on FPGA

对于图像的多尺度特征点提取及匹配存在提取特征有限、算法复杂度高、处理速度慢的问题,提出了一种利用多尺度Harris角点方法提取图像中物体边界特征点,采用二维高斯拉普拉斯算子检测图像斑点信号,并基于SIFT特征描绘子计算特征描绘子相似度来完成匹配的方法。该方法充分利用FPGA的并行性特点,通过降低多尺度Harris角点检测、斑点提取和SIFT特征描绘子生成算法的复杂度,在FPGA中实现了图像多尺度特征点的快速提取及匹配。通过对多组图片进行对比测试验证,该方法相对于软件实现方法具有处理速度快,而图像主要特征点不丢失的特点。

In view of some problem such as limited exacted feature, high algorithm complexity and slow processing speed existed in muhi-scale feature points extraction and matching of images, a method is presented, which combines the techniques of capturing feature points of intersection of object boundary in an image by using multi-scale Harris corner detection algorithm and detecting 2D image spot signal by Laplace of Gaussian, and completing matching of feature points by similarities of SIFT feature descriptor. The algorithm complexities of multi-scale Harris corner detection algorithm, spots extraction method and SIFT feature descriptor＇ s generation method has been greatly reduced to realize in FPGA to achieve fast speed based on parallel characteristics of FPGA. The experiments demonstrate that these techniques can achieve faster speed than the software implementation method with few image feature points loss.