基于NCC的改进立体匹配算法

Improved stereo matching algorithm based on NCC

在双目立体视觉系统中,图像匹配是关键步骤之一。在众多匹配算法中,归一化互相关(NCC)算法由于具有精度高、鲁棒性强等优点得到广泛应用,但其计算量大、运算速度较慢,使其难以在线应用。为此,本文提出一种改进的NCC立体匹配算法,通过引入积分图像和平方积分图像,将矩形窗口区域像素求和运算转化为四个像素点值的简单相加减,同时剔除基准图像中无法匹配区域以减小搜索范围,使计算复杂度得到简化,计算量大为降低。实验证明,改进后的NCC算法在保证匹配质量的基础上,执行速度得到显著提高,利于在线应用。

Image matching is a key step in the binocular stereo vision system. Many methods were proposed to process the problem, among which the normalized cross correlation algorithm is usually used. It has the advantages of high robust and precision,but its computing process is so complicated that difficult to apply on line. Therefore, the integral image and squared integral image are introduced, an improved NCC algorithm is proposed. By using it, the summation operation of pixels′ value in a certain region will be reduce greatly, that is, it only requires to operate additions and subtractions of four pixels′ value. Simultaneously, mismatch areas in the source image are eliminated to decrease the searching range. The improved NCC algorithm not only guarantees the match quality, but also improves the speed of execution greatly. Finally, the simulation results demonstrate its validity.