中文说明:应用背景SURF意指 加速的具有鲁棒性的特征,由Bay在2006年首次提出,这项技术可以应用于计算机视觉的物体识别以及3D重构中。SURF算子由SIFT算子改进而来,一般来说,标准的SURF算子比SIFT算子快好几倍,并且在多幅图片下具有更好的鲁棒性。SURF最大的特征在于采用了harr特征以及积分图像integral image的概念,这大大加快了程序的运行时间。关键技术 surf提出算法参见http://www.vision.ee.ethz.ch/~surf/papers.html 有paper下载地址。 1、提取特征点 2、提取特征描述符 1. 特征点的提取 1)利用Hessian矩阵,计算特征值α 其中Lxx(x, σ)是高斯滤波后图像g(σ)的在x方向的二阶导数,其他的Lyy(x, σ)、Lxy(x, σ)都是g(σ)的二阶导数。 为了减小计算量,原文使用了一个简单的方法,并利用了积分图像的优势(大大的减少计算量),方法其实很简单就是在模糊的基础上将原本的模块近似下。 总所周知,一般计算图像的二阶导时,利用下面的公式d2f(x)/dx2=(f(x+1)-f(x))-(f(x)-f(x-1))=-2*f(x)+f(x+1)+f(x-1)。但是f(x)=g(h(x))【h(x)为图像的灰度值,f(x) 是将h(x)高斯滤波处理的灰度函数 】
English Description:
Application backgroundSURF meansThe robustness of the proposed technique is first proposed by Bay in 2006, which can be applied to object recognition and 3D reconstruction in computer vision. SURF operator is improved by SIFT operator. In general, the standard SURF operator is better than the SIFT operator, and it has better robustness in many pictures. The biggest feature of SURF is the use of Harr features and integral image integralImage concept, which greatly speeds up the running time of the program.Key TechnologySurf proposed algorithm see http://www.vision.ee.ethz.ch/~surf/papers.html have paper Download address.1, extracting feature points2, extract feature descriptor1 feature points extraction1) using the Hessian matrix, the eigenvalue is calculated.The LXX (x, sigma) is the Gaussian filtered image g (sigma) in the X direction of the second order derivative and other Lyy (x, sigma), Lxy (x, sigma) is g (sigma) of the second derivative