（为确保整篇译文的阅读流畅性和连续性，本文主要采用意译，一些和SURF算法关系不大的句子和单词可能不会翻译出来。为帮助理解，可能会增加注释和C++实现代码。）
（原论文下载链接：https://pan.baidu.com/s/1c2auCFm）
原作者信息：
摘要
本文介绍了一种全新的尺度不变、旋转不变的特征点检测和描述算法，英文名为Speed-Up Robust Features，中文可理解为“稳定特征加速算法”，以下简称“SURF”。在特征的重复性、显著性、鲁棒性等关键指标上，相比以往的方案，SURF毫不逊色；在计算和匹配速度方面，则远超前者。
SURF的计算加速能力主要来自以下几个因素：1.基于积分图像，进行卷积计算；2.综合了当前各种主流检测子和描述子的优点（例如，基于Hessain矩阵的检测方法，基于梯度分布的描述子），把这些算法的核心思想提炼出来，加以简化，由此设计出一套全新的检测、描述和匹配机制。
本文详细说明了SURF算法的检测和描述机制，并探究了算法中关键参数的调整效果。在本文结束部分，介绍了SURF算法在相机标定和物体识别这两个挑战性领域的应用。实验表明，SURF算法在计算机视觉各个分支领域具有广泛实用性。
关键词：兴趣点，局部特征，特征描述，相机标定，物体识别
1. 引言
计算机视觉应用（如图像配准、相机标定、物体识别、图像复原等）的一个关键任务是在同一场景或同一物体的两幅图像中寻找匹配特征点。
寻找离散图像匹配特征点的过程可分为以下三个步骤：
(1)在图像灰度突变区域（如角点、blob、T形交点）选择“兴趣点”。这需要首先定义一个兴趣点检测子。检测子最重要的属性是可重复性，即检测子在不同视觉条件下能够找到相同兴趣点的能力的稳定性。
(2)将每个兴趣点的领域特征用一个特征向量表示——即“描述子”。描述子必须有较高的区分度，且对噪点、位移、几何形变和光照变化具有鲁棒性。
(3)最后，在不同图像之间寻找欧拉距离最短的特征向量，欧拉距离小于阈值的向量所在像素即为匹配点。特征描述子的维数直接影响到匹配所需要的计算时间。通常情况下，维数越低，匹配计算时间越短，但特征向量的可区分度也随之降低。原文:
Speeded-UpRobust Features (SURF)
Abstract
Thisarticle presents a novel scale- and rotation-invariant detector and descriptor,coined SURF (Speeded-Up Robust Features).
SURFapproximates or even outperforms previously proposed schemes with respect torepeatability, distinctiveness, and robustness, yet can be computed andcompared much faster.
Thisis achieved by relying on integral images for image convolutions; by buildingon the strengths of the leading existing detectors and descriptors(specifically, using a Hessian matrix-based measure for the detector, and adistribution-based descriptor); and by simplifying these methods to theessential. This leads to a combination of novel detection, description, andmatching steps.
Thepaper encompasses a detailed description of the detector and descriptor andthen explores the effect of the most important parameters. We conclude thearticle with SURF’s application to two challenging, yet converse goals: cameracalibration as a special case of image registration, and object recognition.Our experiments underline SURF’s usefulness in a broad range of topics in computervision.
Keywords:
interestpoints, local features, feature description, camera calibration, objectrecognition
1. Introduction
The task of finding point correspondences between twoimages of the same scene or object is part of many computervision applications. Image registration, camera calibration,object recognition, and image retrieval are just a few.
The search for discrete image point correspondences canbe divided into three main steps. First, ‘interest points’are selected at distinctive locations in the image, such ascorners, blobs, and T-junctions. The most valuable propertyof an interest point detector is its repeatability. Therepeatability expresses the reliability of a detector for findingthe same physical interest points under different viewingconditions. Next, the neighbourhood of every interestpoint is represented by a feature vector. This descriptor hasto be distinctive and at the same time robust to noise, detectiondisplacements and geometric and photometric deformations.Finally, the descriptor vectors are matched betweendifferent images. The matching is based on a distancebetween the vectors, e.g. the Mahalanobis or Euclidean distance.The dimension of the descriptor has a direct impacton the time this takes, and less dimensions are desirable forfast interest point matching. However, lower dimensionalfeature vectors are in general less distinctive than theirhigh-dimensional counterparts.