姓名：鄒富
學(xué)號(hào)：20021211160
【嵌牛導(dǎo)讀】特征提取中各個(gè)算法性能的比較
【嵌牛鼻子】特征提取算法
【嵌牛正文】
轉(zhuǎn)載自：https://blog.csdn.net/h1yupyp/article/details/81042414
在前面的blog中炫掐，我們已經(jīng)講了SIFT的原理，這里我們?cè)僭敿?xì)講解SIFT的變體：PCA-SIFT和GLOH。– Scale invariant feature transform (SIFT): Lowe, 2004.– PCA-SIFT: SIFT: Ke and Sukthankar 2004 Ke and Sukthankar, 2004.– Gradient location-orientation histogram (GLOH): Mikolajczyk and Schmid 2005– SURF(Speeded Up Robust Features), Bay, 2006回顧前面講過的SIFT算法材泄，可以很好地應(yīng)對(duì)旋轉(zhuǎn)和尺度不變，光強(qiáng)不變，位置遮擋不變（http://blog.csdn.net/abcjennifer/article/details/7639681）蜡秽，其過程分為四步：– Detection of scale-space extreme 構(gòu)建尺度空間– Accurate keypoint localization 關(guān)鍵點(diǎn)檢測(cè)– Orientation assignment 指定方向– The local image descriptor 局部圖像描述子David G. Lowe, "Distinctive image features from scale-invariant keypoints," International Journal of Computer Vision, 60, 2 (2004), pp. 91-110/***************************************************PCA-SIFT*****************************************************/PCA(Principle component analysis) SIFT 描述子將在所有描述子中提取出更有區(qū)分度，更robust to image deformations的特征缆镣。其方法：– 在第四步中芽突，不用原先的4*4*8個(gè)描述子，而是在41*41的圖像塊上計(jì)算39*39*2（x,y方向）個(gè)梯度導(dǎo)數(shù)董瞻，然后使用PCA將得到的3042維向量降到36維寞蚌。Y. Ke and R. Sukthankar, “PCA-SIFT: A More Distinctive Representation for Local Image 15 Descriptors,” Computer Vision and Pattern Recognition, 2004.
當(dāng)然，上圖只是PCA-SIFT作者的一面之詞，Mikolajczyk and Schmid（2005）的描述子測(cè)評(píng)顯示還是SIFT比較靠譜挟秤。/***************************************************GLOH*****************************************************/Mikolajczyk and Schmid（2005）提出了一種SIFT變體的描述子壹哺，使用對(duì)數(shù)極坐標(biāo)分級(jí)結(jié)構(gòu)替代Lowe（2004）使用的4象限∩反希空間上取半徑6斗躏，11逝慧，15昔脯，角度上分八個(gè)區(qū)間（除中間區(qū)域）,然后將272（17*16）維的histogram在一個(gè)大數(shù)據(jù)庫(kù)上訓(xùn)練，用PCA投影到一個(gè)128維向量笛臣。
K. Mikolajczyk and C. Schmid,“A performance evaluation of local descriptors ,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 27, No. 10, pp. 1615-1630, Oct. 2005/***************************************************SURF*****************************************************/SURF與SIFT稍有不同云稚，-SIFT建立一幅圖像的金字塔，在每一層進(jìn)行高斯濾波并求取圖像差（DOG）進(jìn)行特征點(diǎn)的提取沈堡，而SURF用的是hessian matrix黑森矩陣静陈。-SIFT特征建立圖像金字塔處理尺度不變特性，而SURF特征將高斯核近似為一個(gè)方波濾波诞丽，SURF金字塔僅僅用來作特征點(diǎn)的檢測(cè)鲸拥。下文來自《A Comparison of SIFT, PCA-SIFT and SURF》SIFT and SURF algorithms employ slightly different ways of detecting features [9]. SIFT builds an imagepyramids, filtering each layer with Gaussians of increasing sigma values and taking the difference. On theother hand, SURF creates a “stack” without 2:1 down sampling for higher levels in the pyramid resultingin images of the same resolution [9]. Due to the use of integral images, SURF filters the stack using a boxfilter approximation of second-order Gaussian partial derivatives, since integral images allow thecomputation of rectangular box filters in near constant time [3].?In keypoint matching step, the nearest neighbor is defined as the keypoint with minimum Euclideandistance for the invariant descriptor vector. Lowe used a more effective measurement that obtained bycomparing the distance of the closest neighbor to that second-closest neighbor [1] so the author of thispaper decided to choose 0.5 as distance ratio like Lowe did in SIFT.Herbert Bay, Andreas Ess, Tinne Tuytelaars, Luc Van Gool?"SURF: Speeded Up Robust Features", Computer Vision and Image Understanding (CVIU), Vol. 110, No. 3, pp. 346--359, 2008ftp://ftp.vision.ee.ethz.ch/publications/articles/eth_biwi_00517.pdf/***************************************************COMPARISON*****************************************************/論文：A comparison of SIFT, PCA-SIFT and SURF 對(duì)三種方法給出了性能上的比較，源圖片來源于Graffiti dataset僧免，對(duì)原圖像進(jìn)行尺度刑赶、旋轉(zhuǎn)、模糊懂衩、亮度變化撞叨、仿射變換等變化后，再與原圖像進(jìn)行匹配浊洞，統(tǒng)計(jì)匹配的效果牵敷。效果以可重復(fù)出現(xiàn)性為評(píng)價(jià)指標(biāo)。? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 對(duì)以上三種方法進(jìn)行比較：
由此可見法希，SIFT在尺度和旋轉(zhuǎn)變換的情況下效果最好枷餐，SURF在亮度變化下匹配效果最好，在模糊方面優(yōu)于SIFT苫亦，而尺度和旋轉(zhuǎn)的變化不及SIFT毛肋，旋轉(zhuǎn)不變上比SIFT差很多。速度上看著觉，SURF是SIFT速度的3倍村生。
采用最近鄰作為匹配策略的特征描述子性能測(cè)評(píng)結(jié)果：
Reference:http://blog.csdn.net/abcjennifer/article/details/7365651http://www.cscjournals.org/csc/manuscript/Journals/IJIP/volume3/Issue4/IJIP-51.pdfhttp://www.cnblogs.com/mysunnyday/archive/2011/08/31/2160298.htmlhttp://140.115.156.251/vclab/teacher/2011AIP/Feature%20Detection%20and%20Matching%20(Part%20II).pdf
