計(jì)算機(jī)視覺(jué)的一些測(cè)試數(shù)據(jù)集和源碼站點(diǎn)

圖像處理（54） 

計(jì)算機(jī)視覺(jué)（72） 

機(jī)器學(xué)習(xí)（45） 

zouxy09@QQ.com

http://blog.csdn.net/zouxy09

轉(zhuǎn)自：http://blog.csdn.net/zhubenfulovepoem/article/details/7191794

       以下是computer vision：algorithm and application計(jì)算機(jī)視覺(jué)算法與應(yīng)用這本書(shū)中附錄里的關(guān)于計(jì)算機(jī)視覺(jué)的一些測(cè)試數(shù)據(jù)集和源碼站點(diǎn)，我整理了下，加了點(diǎn)中文注解。

ComputerVision:

Algorithms and Applications

Richard Szeliski

在http://szeliski.org/Book包含了更新的數(shù)據(jù)集和軟件，請(qǐng)同樣訪問(wèn)他。

C.1 數(shù)據(jù)集

一個(gè)關(guān)鍵就是用富有挑戰(zhàn)和典型的數(shù)據(jù)集來(lái)測(cè)試你算法的可靠性。當(dāng)有背景或者他人的結(jié)果是可行的,這種測(cè)試可能甚至包含更多的信息(和質(zhì)量更好)。

經(jīng)過(guò)這些年，大量的數(shù)據(jù)集已經(jīng)被提出來(lái)用于測(cè)試和評(píng)估計(jì)算機(jī)視覺(jué)算法。許多這些數(shù)據(jù)集和軟件被編入了計(jì)算機(jī)視覺(jué)的主頁(yè)。一些更新的網(wǎng)址，像CVonline

(http://homepages.inf.ed.ac.uk/rbf/CVonline ), VisionBib.Com (http://datasets.visionbib.com/ ), and Computer Vision online (http://computervisiononline.com/ ), 有更多最新的數(shù)據(jù)集和軟件。

下面，我列出了一些用的最多的數(shù)據(jù)集，我將它們讓章節(jié)排列以便它們聯(lián)系更緊密。

第二章：圖像信息

CUReT: Columbia-Utrecht 反射率和紋理數(shù)據(jù)庫(kù)Re?ectance and TextureDatabase,http://www1.cs.columbia.edu/CAVE/software/curet/ (Dana, van Ginneken, Nayaret al. 1999).

Middlebury Color Datasets:不同攝像機(jī)拍攝的圖像，注冊(cè)后用于研究不同的攝像機(jī)怎么改變色域和彩色registeredcolor images taken by different cameras to study how they transform gamuts andcolors, http://vision.middlebury.edu/color/data/ Chakrabarti, Scharstein, and Zickler 2009).

第三章：圖像處理

Middlebury test datasets forevaluating MRF minimization/inference algorithms評(píng)估隱馬爾科夫隨機(jī)場(chǎng)最小化和推斷算法,

http://vision.middlebury.edu/MRF/results/ (Szeliski, Zabih, Scharstein et al. 2008).

第四章：特征檢測(cè)和匹配

Af?ne Covariant Featuresdatabase（反射協(xié)變的特征數(shù)據(jù)集） for evaluating feature detector and descriptor matching quality andrepeatability（評(píng)估特征檢測(cè)和描述匹配的質(zhì)量和定位精度）, http://www.robots.ox.ac.uk/~vgg/research/affine/

(Miko-lajczyk and Schmid 2005;Mikolajczyk, Tuytelaars, Schmid et al. 2005).

Database of matched imagepatches for learning （圖像斑塊匹配學(xué)習(xí)數(shù)據(jù)庫(kù)）and feature descriptor evaluation（特征描述評(píng)估數(shù)據(jù)庫(kù)）,

http://cvlab.epfl.ch/~brown/patchdata/patchdata.html

(Winder and Brown 2007;Hua,Brown, and Winder 2007).

第五章;分割

BerkeleySegmentation Dataset（分割數(shù)據(jù)庫(kù)） and Benchmark of 1000 images labeled by 30 humans,（30個(gè)人標(biāo)記的1000副基準(zhǔn)圖像）along with an evaluation,http://www.eecs.berkeley.edu/Research/PRojects/CS/vision/grouping/segbench/ (Martin, Fowlkes, Tal et al.2001).

Weizmann segmentationevaluation database of 100 grayscale images with ground truth segmentations,

http://www.wisdom.weizmann.ac.il/~vision/Seg EvaluationDB/index.html

(Alpert, Galun, Basri et al. 2007).

第八章：稠密運(yùn)動(dòng)估計(jì)

TheMiddlebury optic ?ow evaluation（光流評(píng)估） Web site,http://vision.middlebury.edu/flow/data/

(Baker,Scharstein, Lewis et al. 2009).

The Human-Assisted MotionAnnotation database,（人類(lèi)輔助運(yùn)動(dòng)數(shù)據(jù)庫(kù)）

http://people.csail.mit.edu/celiu/motionAnnotation/ (Liu, Freeman, Adelson etal. 2008)

第十章：計(jì)算機(jī)攝像學(xué)

High DynamicRange radiance（輻射）maps, http://www.debevec.org/Research/HDR/

(De-bevecand Malik 1997).

Alpha matting evaluation Website, http://alphamatting.com/ (Rhemann, Rother, Wang

et al. 2009).

第十一章：Stereo correspondence立體對(duì)應(yīng)

Middlebury Stereo Datasets andEvaluation, http://vision.middlebury.edu/stereo/(Scharstein

and Szeliski 2002).

StereoClassi?cation（立體分類(lèi)） and Performance Evaluation（性能評(píng)估） of different aggregation（聚類(lèi)） costs for stereo matching（立體匹配）,http://www.vision.deis.unibo.it/spe/SPEHome.aspx (Tombari, Mat-

toccia, Di Stefano et al.2008).

Middlebury Multi-View StereoDatasets,

http://vision.middlebury.edu/mview/data/ (Seitz,Curless, Diebel etal. 2006).

Multi-view and Oxford Collegesbuilding reconstructions,

http://www.robots.ox.ac.uk/~vgg/data/data-mview.html .

Multi-View Stereo Datasets, http://cvlab.epfl.ch/data/strechamvs/ (Strecha, Fransens,

and Van Gool 2006).

Multi-View Evaluation, http://cvlab.epfl.ch/~strecha/multiview/ (Strecha, von Hansen,

Van Gool et al. 2008).

第十二章：3D重建

HumanEva: synchronized video（同步視頻） and motion capture （動(dòng)作捕捉）dataset for evaluation ofarticulated human motion, http://vision.cs.brown.edu/humaneva/ Sigal, Balan, and Black 2010).

第十三章：圖像渲染

The (New) Stanford Light FieldArchive, http://lightfield.stanford.edu/

(Wilburn, Joshi,Vaish et al.2005).

Virtual Viewpoint Video:multi-viewpoint video with per-frame depth maps,

http://research.microsoft.com/en-us/um/redmond/groups/ivm/vvv/ (Zitnick, Kang, Uytten-

daele et al. 2004).

第十四章：識(shí)別

查找一系列的視覺(jué)識(shí)別數(shù)據(jù)庫(kù)，在表14.1–14.2.除了那些，這里還有：

Buffy pose classes, http://www.robots.ox.ac.uk/~vgg/data/ buffy pose classes/ andBuffy

stickmen V2.1, http://www.robots.ox.ac.uk/~vgg/data/stickmen/index.html (Ferrari,Marin-

Jimenez, and Zisserman 2009;Eichner and Ferrari 2009).

H3D database of pose/jointannotated photographs of humans,

http://www.eecs.berkeley.edu/~lbourdev/h3d/ (Bourdev and Malik 2009).

Action Recognition Datasets,http://www.cs.berkeley.edu/projects/vision/action, has point-

ers toseveral datasets for action and activity recognition, as well as some papers.（有一些關(guān)于人活動(dòng)和運(yùn)動(dòng)的數(shù)據(jù)庫(kù)和論文） The humanaction database athttp://www.nada.kth.se/cvap/actions/ 包含更多的行動(dòng)序列。

C.2 軟件資源

一個(gè)對(duì)于計(jì)算機(jī)視覺(jué)算法最好的資源就是開(kāi)源視覺(jué)圖像庫(kù)（OpenCV）(http://opencv.willowgarage.com/wiki/),他有在intel的Gary Bradski和他的同事開(kāi)發(fā)，現(xiàn)在由Willow Garage (Bradsky and Kaehler 2008)維護(hù)和擴(kuò)展。一部分可利用的函數(shù)在http://opencv.willowgarage.com/documentation/cpp/中：

圖像處理和變換 (濾波，形態(tài)學(xué)，金字塔);

圖像幾何學(xué)的變換 (旋轉(zhuǎn)，改變大小);

混合圖像變換 (傅里葉變換，距離變換);

直方圖;

分割 (分水嶺, mean shift);

特征檢測(cè) (Canny, Harris, Hough, MSER, SURF);

運(yùn)動(dòng)分析和物體分析 (Lucas–Kanade, mean shift);

相機(jī)矯正和3D重建

機(jī)器學(xué)習(xí) (k nearest neighbors, 支持向量機(jī), 決策樹(shù), boost-

ing, 隨機(jī)樹(shù), expectation-maximization, 和神經(jīng)網(wǎng)絡(luò)).

Intel的Performance Primitives (IPP)library, http://software.intel.com/en-us/intel-ipp/，包含

各種各樣的圖像處理任務(wù)的最佳優(yōu)化代碼，許多opencv中的例子利用了這個(gè)庫(kù)，加入他安裝了，程序運(yùn)行得更快。依據(jù)功能，他和Opencv有很多相同的運(yùn)算處理，并且加上了額外的庫(kù)針對(duì)圖像視頻壓縮，信號(hào)語(yǔ)音處理和矩陣代數(shù)。

MTALAB中的Image Processing Toolbox圖像處理工具，http://www.mathworks.com/products/image/，包含常規(guī)的處理，空域變換（旋轉(zhuǎn)，改變大?。?，常規(guī)正交，圖像分析和統(tǒng)計(jì)學(xué)（變邊緣，哈弗變換），圖像增強(qiáng)（自適應(yīng)直方圖均衡，中值濾波），圖像恢復(fù)（去模糊），線性濾波（卷積），圖像變換（傅里葉，離散余弦變換）和形態(tài)學(xué)操作（連通域和距離變換）

兩個(gè)比較舊的庫(kù)，它們沒(méi)有被發(fā)展，但是包含了一些的有用的常規(guī)操作：

VXL (C++Libraries for Computer Vision Research and Implemen-tation,http://vxl.sourceforge.net/)

LTI-Lib 2 (http://www.ie.itcr.ac.cr/palvarado/ltilib-2/homepage/ ).

圖像編輯和視圖包，例如Windows Live Photo Gallery, iPhoto, Picasa,GIMP, 和 IrfanView，它們對(duì)執(zhí)行這些處理非常有用：常規(guī)處理任務(wù)，格式轉(zhuǎn)換，觀測(cè)你的結(jié)果。它們同樣可以用于對(duì)圖像處理算法有趣的實(shí)現(xiàn)參考，例如色調(diào)調(diào)整和去噪。

這里他也有一些軟件包和基礎(chǔ)框架對(duì)你建一個(gè)實(shí)時(shí)視頻處理的DEMOS很有用，Vision on Tap(http://www.visionontap.com/ )提供一個(gè)可以實(shí)時(shí)處理你的網(wǎng)絡(luò)攝像頭的網(wǎng)頁(yè)服務(wù)(Chiu and Raskar 2009）。Video-Man (VideoManager,http://videomanlib.sourceforge.net/處理實(shí)時(shí)的基于視頻的DEMOS和應(yīng)用非常有用，你也可以用MATLAB中的imread直接從任何URl（例如網(wǎng)絡(luò)攝像頭）中讀取視頻。

下面，我列出了一些額外的網(wǎng)絡(luò)資源，讓章節(jié)排列以便它們看起來(lái)聯(lián)系更緊密：

第三章:圖像處理

matlabPyrTools—MATLAB 下的源碼對(duì)于拉普拉斯變換，金字塔, QMF/小波, 和

steerable pyramids, http://www.cns.nyu.edu/~lcv/software.php (Simoncelli and Adel-

son 1990a; Simoncelli,Freeman, Adelson et al. 1992).

BLS-GSM 圖像去噪, http://decsai.ugr.es/~javier/denoise/ (Portilla, Strela,Wain-

wright et al. 2003).

Fast bilateral ?ltering code（快速雙邊濾波）, http://people.csail.mit.edu/jiawen/#code(Chen, Paris, and Durand 2007).

C++ implementation of the fastdistance transform algorithm,

http://people.cs.uchicago.edu/~pff/dt/ (Felzenszwalb andHuttenlocher 2004a).

GREYC’s Magic Image Converter,including image restoration software using regularization and anisotropicdiffusion, http://gmic.sourceforge.net/gimp.shtml(Tschumperl′ e and Deriche 2005).

第四章：圖像特征檢測(cè)和匹配

VLFeat, 一個(gè)開(kāi)放便捷的計(jì)算機(jī)視覺(jué)算法庫(kù)

http://vlfeat.org/ (Vedaldi and Fulkerson 2008).

SiftGPU: A GPU Implementationof Scale Invariant Feature Transform (SIFT),

GPU實(shí)現(xiàn)的尺度特征性變換

http://www.cs.unc.edu/~ccwu/siftgpu/ (Wu 2010).

SURF: Speeded Up RobustFeatures, http://www.vision.ee.ethz.ch/~surf/

(Bay, Tuyte-laars, and VanGool 2006).

FAST corner detection, http://mi.eng.cam.ac.uk/~er258/work/fast.html

(Rosten and Drum-mond 2005, 2006).

linux binaries for af?neregion detectors and descriptors, as well as MATLAB ?les to

compute repeatability andmatching scores,

http://www.robots.ox.ac.uk/~vgg/research/affine/

Kanade–Lucas–Tomasi featuretrackers: KLT, http://www.ces.clemson.edu/~stb/klt/ (Shi and Tomasi 1994);

GPU-KLT, http://cs.unc.edu/~cmzach/opensource.html (Zach,Gallup, and Frahm2008); Lucas–Kanade 20 Years On, http://www.ri.cmu.edu/projects/project 515.html (Baker and Matthews 2004).

第五章：分割

高效的基于圖形的分割http://people.cs.uchicago.edu/~pff/segment

(Felzenszwalb and Huttenlocher2004b).

EDISON, 邊緣檢測(cè)和圖像追蹤,

http://coewww.rutgers.edu/riul/research/code/EDISON/

(Meer and Georgescu 2001; Comaniciu and Meer2002).

Normalized cuts segmentationincluding intervening contours,

http://www.cis.upenn.edu/~jshi/software/

(Shi and Malik 2000; Malik,Belongie, Leung et al. 2001).

Segmentation by weightedaggregation (SWA),利用加權(quán)集合的分割

http://www.cs.weizmann.ac.il/~vision/SWA (Alpert, Galun, Basri et al.2007).

第六章：基于特征的對(duì)齊和校準(zhǔn)

Non-iterative PnP algorithm,（非迭代PnP算法）

http://cvlab.ep?.ch/software/EPnP (Moreno-Noguer, Lep-etit, and Fua 2007).

Tsai Camera Calibration（相機(jī)矯正） Software,

http://www-2.cs.cmu.edu/~rgw/TsaiCode.html (Tsai 1987).

Easy CameraCalibration Toolkit,（簡(jiǎn)易相機(jī)校準(zhǔn)工具包）http://research.microsoft.com/en-us/um/people/zhang/ Calib/ (Zhang 2000).

Camera Calibration Toolbox forMATLAB,

http://www.vision.caltech.edu/bouguetj/calib doc/ ; a C version is included in OpenCV.

MATLAB functions for multipleview geometry,

http://www.robots.ox.ac.uk/~vgg/hzbook/code/ (Hartley and Zisserman2004).

第七章：運(yùn)動(dòng)重建

SBA: A generic sparse bundle(稀疏束) adjustment C/C++ package basedon the Levenberg–

Marquardt algorithm, http://www.ics.forth.gr/~lourakis/sba/ (Lourakis and Argyros 2009).

Simple sparse bundleadjustment (SSBA), http://cs.unc.edu/~cmzach/opensource.html .

Bundler, structure from motionfor unordered image collections(無(wú)序圖像集),

http://phototour.cs.washington.edu/bundler/ (Snavely, Seitz, and Szeliski 2006).

第八章:稠密運(yùn)動(dòng)估計(jì)

光流, http://www.cs.brown.edu/~black/code.html (Black and Anan-

dan 1996).

Optical ?ow（光流） using total variation（全變量差） and conjugate gradientdescent（共軛梯度下降）, http://people.csail.mit.edu/celiu/OpticalFlow/ (Liu 2009).

TV-L1 optical ?ow on the GPU, http://cs.unc.edu/~cmzach/opensource.html

(Zach,Pock, and Bischof2007a).

elastix: atoolbox for rigid（剛性） and nonrigid（非剛性） registration of images（配準(zhǔn)圖像）, http://elastix.isi.uu.nl/ (Klein, Staring, and Pluim 2007).

Deformable image registration（可變形的配準(zhǔn)圖像） using discreteoptimization（離散最優(yōu)化）, http://www.mrf-registration.net/deformable/index.html

(Glocker, Komodakis, Tziritas et al. 2008).

第九章：圖像縫合

Microsoft Research ImageCompositing Editor for stitching images,（圖像拼接，圖像合成）

http://research.microsoft.com/en-us/um/redmond/groups/ivm/ice/ .

第十章：計(jì)算機(jī)攝影學(xué)

HDRShop software for combiningbracketed exposures（包圍式曝光） into high-dynamic range radiance images, http://projects.ict.usc.edu/graphics/HDRShop/.

Super-resolution（超分辨率） code,

http://www.robots.ox.ac.uk/~vgg/software/SR/ (Pickup 2007;Pickup, Capel,Roberts et al. 2007, 2009).

第十一章：立體對(duì)應(yīng)

StereoMatcher, standalone C++stereo matching code,

http://vision.middlebury.edu/stereo/code/ (Scharstein and Szeliski2002).

Patch-based multi-view stereosoftware (PMVS Version 2),

http://grail.cs.washington.edu/software/pmvs/ (Furukawa and Ponce 2011).

第十二章：3D重建

Scanalyze: a system foraligning and merging range data,

http://graphics.stanford.edu/software/scanalyze/ (Curless and Levoy 1996).

MeshLab: software forprocessing, editing, and visualizing unstructured 3D triangular

meshes, http://meshlab.sourceforge.net/.

VRML viewers (various) arealso a good way to visualize texture-mapped 3D models.

節(jié) 12.6.4: Whole body modeling andtracking（全身建模和追蹤）

Bayesian 3D person tracking（貝葉斯3D人體追蹤）,http://www.cs.brown.edu/~black/code.html (Sidenbladh,Black, and Fleet2000; Sidenbladh and Black 2003).

HumanEva: baseline code forthe tracking of articulated human motion,

http://vision.cs.brown.edu/humaneva/ (Sigal, Balan, and Black 2010).

節(jié) 14.1.1: Face detection（人臉檢測(cè)）

Sample face detection code andevaluation tools,

http://vision.ai.uiuc.edu/mhyang/face-detection-survey.html.

節(jié) 14.1.2: Pedestrian detection（行人追蹤）

A simple object detector withboosting,

http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/boosting.html

(Hastie, Tibshirani, and Friedman 2001;Torralba, Murphy, and Freeman 2007).

Discriminatively（有區(qū)別） trained deformable（可變形） part models,http://people.cs.uchicago.edu/~pff/latent/ (Felzenszwalb, Girshick,McAllester et al. 2010).

Upper-body detector（上身檢測(cè)）,

http://www.robots.ox.ac.uk/~vgg/software/UpperBody/ (Ferrari,Marin-Jimenez, andZisserman 2008).

2D articulated human poseestimation software,

http://www.vision.ee.ethz.ch/~calvin/articulated_human_pose_estimation_code/ (Eichner and Ferrari 2009).

節(jié) 14.2.2: Active appearance and 3Dshape models

AAMtools: An active appearancemodeling toolbox,

http://cvsp.cs.ntua.gr/software/AAMtools/ (Papandreou and Maragos2008).

節(jié) 14.3: Instance recognition

FASTANN and FASTCLUSTER forapproximate k-means (AKM),

http://www.robots.ox.ac.uk/~vgg/software/ (Philbin, Chum, Isard et al. 2007).

Feature matching using fastapproximate nearest neighbors,

http://people.cs.ubc.ca/~mariusm/index.php/FLANN/FLANN (Muja and Lowe 2009).

節(jié) 14.4.1: Bag of Words(詞袋)

Two bag of words classi?ers, http://people.csail.mit.edu/fergus/iccv2005/bagwords.html

(Fei-Fei and Perona 2005;Sivic, Russell, Efros et al. 2005).

Bag of features andhierarchical（分層） k-means, http://www.vlfeat.org/ (Nist′ er and Stew′enius2006; Nowak, Jurie, and Triggs 2006).

節(jié) 14.4.2: Part-based models

A simple parts and structureobject detector,

http://people.csail.mit.edu/fergus/iccv2005/partsstructure.html

(Fischler and Elschlager 1973; Felzenszwalband Huttenlocher 2005).

節(jié) 14.5.1: Machine learning software

Support vector machines (SVM)software (

http://www.support-vector-machines.org/SVM soft.html )

包含很多支持向量機(jī)的庫(kù),

SVMlight http://svmlight.joachims.org/ ;

LIBSVM, http://www.csie.ntu.edu.tw/~cjlin/libsvm/(Fan, Chen,and Lin 2005);

LIBLINEAR, http://www.csie.ntu.edu.tw/~cjlin/liblinear/ (Fan,Chang, Hsieh et al.2008).

Kernel Machines: links to SVM,Gaussian processes, boosting, and other machine

learning algorithms, http://www.kernel-machines.org/software .

Multiple kernels for imageclassi?cation,

http://www.robots.ox.ac.uk/~vgg/software/MKL

(Varma and Ray 2007; Vedaldi, Gulshan, Varmaet al. 2009).

附錄 A.1–A.2: Matrix decompositions（矩陣分解） and linear least squares（線性最小乘）

BLAS (BasicLinear Algebra Subprograms基本線性代數(shù)子程序),

http://www.netlib.org/blas/ (Blackford,Demmel, Dongarraet al. 2002).

LAPACK (Linear Algebra（線性代數(shù)） PACKage),

http://www.netlib.org/lapack/ (Anderson, Bai,Bischof etal. 1999).

GotoBLAS, http://www.tacc.utexas.edu/tacc-projects/.

ATLAS (Automatically TunedLinear Algebra Software),

http://math-atlas.sourceforge.net/ (Demmel, Dongarra, Eijkhoutet al. 2005).

Intel Math Kernel Library(MKL), http://software.intel.com/en-us/intel-mkl/.

AMD CoreMath Library (ACML),

http://developer.amd.com/cpu/Libraries/acml/Pages/default.aspx .

Robust PCA code（魯棒主成分分析）, http://www.salle.url.edu/~ftorre/papers/rpca2.html

(De la Torre and Black 2003).

Appendix A.3: Non-linear leastsquares非線性最小二乘

MINPACK, http://www.netlib.org/minpack/.

levmar: Levenberg–Marquardtnonlinear least squares algorithms, 非線性最小二乘

http://www.ics.forth.gr/~lourakis/levmar/ (Madsen, Nielsen, andTingleff 2004).

附錄 A.4–A.5: Direct（直接） and iterative（迭代） sparse matrix（稀疏矩陣） solvers

SuiteSparse (variousreordering algorithms, 各種各樣的重排算法CHOLMOD) and SuiteSparse QR, http://www.cise.ufl.edu/research/sparse/SuiteSparse/ (Davis 2006, 2008).

PARDISO (iterative and sparsedirect solution), http://www.pardiso-project.org/.

TAUCS (sparse direct,iterative, out of core, preconditioners),

http://www.tau.ac.il/~stoledo/taucs/ .

HSL Mathematical SoftwareLibrary, http://www.hsl.rl.ac.uk/index.html .

Templatesfor the solution of linear systems（線性系統(tǒng)解決問(wèn)題的模板）,http://www.netlib.org/linalg/html templates/Templates.html (Barrett, Berry, Chan et al.1994). Download the PDF for instructions（說(shuō)明） on how to get the software.

ITSOL,MIQR, and other sparsesolvers,

http://www-users.cs.umn.edu/~saad/software/ (Saad 2003).

ILUPACK, http://www-public.tu-bs.de/~bolle/ilupack/ .

附錄 B: Bayesian modeling and inference（貝葉斯建模和推斷）

Middleburysource code for MRF minimization（隱馬爾科夫隨機(jī)場(chǎng)最小化）,http://vision.middlebury.edu/MRF/code/ (Szeliski, Zabih, Scharsteinet al. 2008).

C++ code for ef?cient beliefpropagation for early vision,

http://people.cs.uchicago.edu/~pff/bp/ (Felzenszwalb andHuttenlocher 2006).

FastPD MRF optimization（最優(yōu)化） code,

http://www.csd.uoc.gr/~komod/FastPD (Komodakisand Tziritas2007a; Komodakis, Tziritas, and Paragios 2008)

算法 C.1 Calgorithm for Gaussian random noise generation, using the Box–Mullertransform.

C描述的利用Box–Muller 變換產(chǎn)生高斯隨機(jī)噪聲

double urand()

{

return ((double)rand()) / ((double) RAND MAX);

}

void grand(double& g1, double& g2)

{

#ifndef M_PI

#define M_PI 3.14159265358979323846

#endif // M_PI

double n1 = urand();

double n2 = urand();

double x1 = n1 + (n1 == 0); /* guardagainst log(0) */

double sqlogn1 = sqrt(-2.0 * log (x1));

double angl = (2.0 * M PI) * n2;

g1 = sqlogn1 * cos(angl);

g2 = sqlogn1 * sin(angl);

}

高斯噪聲的產(chǎn)生。許多基本的軟件包產(chǎn)生一些不同的隨機(jī)的噪聲(例如 運(yùn)行在unix上的rand())，但是并不是所有的都有高斯隨機(jī)噪聲發(fā)生器。計(jì)算一個(gè)離散隨機(jī)常量，你可以用Box–Mullertransform (Box and Muller 1958)，他的c代碼在算法C.1中給出了，注意這個(gè)運(yùn)行結(jié)果是返回一對(duì)隨機(jī)變量。相關(guān)的產(chǎn)生高斯隨機(jī)變量的方由Thomas, Luk, Leong et al. (2007)提出。

偽彩色產(chǎn)生。在很多應(yīng)用中，很方便給圖像加上標(biāo)記（或者給圖像特征比如線）。一個(gè)最簡(jiǎn)單的方式就是給不同的標(biāo)記不同的顏色。在我的工作中，我發(fā)現(xiàn)用RGB立體色彩系給不同的標(biāo)記賦予標(biāo)準(zhǔn)均勻的色彩是很方便的。

對(duì)于每一個(gè)（非消極）標(biāo)記值，considerthe bits as being split among the three color channel，例如對(duì)于一個(gè)比特值為9的值，

這個(gè)值可以被標(biāo)記為RGBRGBRGB，獲得三基色中的每一種顏色值后，顛倒比特值，結(jié)果是低位的比特值變化的最快。

實(shí)際上，對(duì)于一個(gè)八比特的顏色通道，這個(gè)比特值的顛倒可以被存在一個(gè)表或者一個(gè)存儲(chǔ)提前計(jì)算好的記錄有由標(biāo)記值向偽彩色的改變的完整表。

圖 8.16 顯示了這樣一個(gè)偽彩色繪制的例子.

GPU實(shí)現(xiàn)

GPU的出現(xiàn)，可以處理像素著色和計(jì)算著色，導(dǎo)致了實(shí)時(shí)應(yīng)用的快速計(jì)算機(jī)視覺(jué)算法的發(fā)展，例如，分割，追蹤，立體和運(yùn)動(dòng)估計(jì)（(Pock, Unger, Cremerset al. 2008; Vineet and Narayanan 2008; Zach,Gallup, and Frahm 2008）。一個(gè)好的資源來(lái)學(xué)習(xí)這些算法就是CVPR 2008 上關(guān)于Visual Computer Visionon GPUs的workshop。

http://www.cs.unc.edu/~jmf/Workshop_on_Computer_Vision_on_GPU.html他的論文可以在CVPR2008的會(huì)議集的DVD中找到。額外的關(guān)于GPU算法資源包括GPGPU網(wǎng)址和小組討論http://gpgpu.org/還有OpenVIDIAWeb site,http://openvidia.sourceforge.net/index.php/OpenVIDIA

C.3 PPT和講稿

正如我在前言中提到的，我希望提供和書(shū)中材料相一致的PPT，直到這些全部準(zhǔn)備好，你最好的方式去看我在華盛頓大學(xué)上課時(shí)的PPT，和一寫(xiě)相關(guān)課程中用到的教案。

這里是一些這樣的課程列表：

UW 455:Undergraduate Computer Vision,

http://www.cs.washington.edu/education/courses/455/.

UW576:Graduate Computer Vision,

http://www.cs.washington.edu/education/courses/576.

StanfordCS233B: Introduction to Computer Vision,

http://vision.stanford.edu/teaching/cs223b/.

MIT6.869: Advances in Computer Vision,

http://people.csail.mit.edu/torralba/courses/6.869/6.869.computervision.htm.

Berkeley CS 280: Computer Vision, http://www.eecs.berkeley.edu/~trevor/CS280.html

UNC COMP776: Computer Vision, http://www.cs.unc.edu/~lazebnik/spring10.

Middlebury CS 453: Computer Vision,

http://www.cs.middlebury.edu/~schar/courses/cs453-s10/.

Related courses have also been taught onthe topic of Computational Photography, e.g.,

CMU 15-463: Computational Photography, http://graphics.cs.cmu.edu/courses/15-463/.

MIT 6.815/6.865: Advanced ComputationalPhotography,

http://stellar.mit.edu/S/course/6/sp09/6.815

Stanford CS 448A: Computational photographyon cell phones,

http://graphics.stanford.edu/courses/cs448a-10/.

SIGGRAPH courses on ComputationalPhotography,

http://web.media.mit.edu/~raskar/photo/.

這里還有一些最好的關(guān)于各種計(jì)算機(jī)視覺(jué)主題的在線講稿，例如：belief propagation and graph cuts，它們?cè)赨W-MSR Course of Vision Algo-rithmshttp://www.cs.washington.edu/education/courses/577/04sp/

C.4 參考文獻(xiàn)：

這本的所有參考文獻(xiàn)在這本書(shū)的網(wǎng)站上，一個(gè)幾乎所有的計(jì)算機(jī)視覺(jué)的出版物都引用的更全面的部分注解書(shū)目由Keith Price維http://iris.usc.edu/Vision-Notes/bibliography/contents.html.

這里還有一個(gè)可搜索的計(jì)算機(jī)圖形學(xué)的參考書(shū)目http://www.siggraph.org/publications/bibliography/另外技術(shù)論文比較好的資源是GoogleScholar 和 CiteSeerX。