Lecture 14: Introduction to RecognitionCS6670: Computer VisionNoah Snavelymountainbuildingtreebannervendorpeoplestreet lampAnnouncements• Final project page up, at– http://www.cs.cornell.edu/courses/cs6670/2009fa/projects/p4/– One person from each team should submit a proposal (to CMS) by next Wednesday at 11:59pm• Project 3: eigenfaces– will be posted on the web soon– Adarsh will capture photos at the end of class– project will include a challenge competitionWhat do we mean by ―object recognition‖?Next 15 slides adapted from Li, Fergus, & Torralba’s excellent short course on category and object recognitionVerification: is that a lamp?Detection: are there people?Identification: is that Potala Palace?Object categorizationmountainbuildingtreebannervendorpeoplestreet lampScene and context categorization• outdoor• city• …Object recognitionIs it really so hard?This is a chairFind the chair in this image Output of normalized correlationObject recognitionIs it really so hard?Find the chair in this image Pretty much garbageSimple template matching is not going to make itObject recognitionIs it really so hard?Find the chair in this image A “popular method is that of template matching, by point to point correlation of a model pattern with the image pattern. These techniques are inadequate for three-dimensional scene analysis for many reasons, such as occlusion, changes in viewing angle, and articulation of parts.” Nivatia & Binford, 1977.Brady, M. J., & Kersten, D. (2003). Bootstrapped learning of novel objects. J Vis, 3(6), 413-422 And it can get a lot harderApplications: Computational photographyApplications: Assisted drivingmetersmetersPedPedCarLane detectionPedestrian and car detection• Collision warning systems with adaptive cruise control, • Lane departure warning systems, • Rear object detection systems,Applications: image searchHow do human do recognition? • We don’t completely know yet• But we have some experimental observations.Observation 1• We can recognize familiar faces even in low-resolution imagesObservation 2:Jim CarreyKevin Costner• High frequency information is not enoughObservation 3:• Negative contrast is difficultObservation 3:Observation 4:• Image Warping is OKThe list goes on• http://web.mit.edu/bcs/sinha/papers/19results_sinha_etal.pdfLet’s start simple• Today– skin detection– eigenfacesFace detection• Do these images contain faces? Where?One simple method: skin detectionSkin pixels have a distinctive range of colors• Corresponds to region(s) in RGB color space– for visualization, only R and G components are shown above skinSkin classifier• A pixel X = (R,G,B) is skin if it is in the skin region• But how to find this region?Skin detectionLearn the skin region from examples• Manually label pixels in one or more ―training images‖ as skin or not skin• Plot the training data in RGB space– skin pixels shown in orange, non-skin pixels shown in blue– some skin pixels may be outside the region, non-skin pixels inside. Why?Skin classifier• Given X = (R,G,B): how to determine if it is skin or not?Skin classification techniquesSkin classifier• Given X = (R,G,B): how to determine if it is skin or not?• Nearest neighbor– find labeled pixel closest to X– choose the label for that pixel• Data modeling– fit a model (curve, surface, or volume) to each class• Probabilistic data modeling– fit a probability model to each classProbabilityBasic probability• X is a random variable• P(X) is the probability that X achieves a certain value•• or • Conditional probability: P(X | Y)– probability of X given that we already know Ycontinuous X discrete Xcalled a PDF-probability distribution/density function-a 2D PDF is a surface, 3D PDF is a volumeProbabilistic skin classificationNow we can model uncertainty• Each pixel has a probability of being skin or not skin–Skin classifier• Given X = (R,G,B): how to determine if it is skin or not?• Choose interpretation of highest probability– set X to be a skin pixel if and only if Where do we get and ?Learning conditional PDF’sWe can calculate P(R | skin) from a set of training images• It is simply a histogram over the pixels in the training images– each bin Ricontains the proportion of skin pixels with color RiThis doesn’t work as well in higher-dimensional spaces. Why not?Approach: fit parametric PDF functions • common choice is rotated Gaussian – center – covariance» orientation, size defined by eigenvecs, eigenvalsLearning conditional PDF’sWe can calculate P(R | skin) from a set of training images• It is simply a histogram over the pixels in the training images– each bin Ricontains the proportion of skin pixels with color RiBut this isn’t quite what we want• Why not? How to determine if a pixel is skin?• We want P(skin | R), not P(R | skin)• How can we get it?Bayes ruleIn terms of our problem:what we measure(likelihood)domain knowledge(prior)what we want(posterior)normalization termThe prior: P(skin)• Could use domain knowledge– P(skin) may be larger if we know the image contains a person– for a portrait, P(skin) may be higher for pixels in the center• Could learn the prior from the training set. How?– P(skin) could be the proportion of skin pixels in training setBayesian estimationBayesian estimation• Goal is to choose the label (skin or ~skin) that maximizes the posterior– this is called Maximum A Posteriori (MAP) estimationlikelihood posterior (unnormalized)0.5• Suppose the prior is uniform: P(skin) = P(~skin) = = minimize probability of misclassification– in this case ,– maximizing the posterior is equivalent to maximizing the likelihood» if and only if – this is called Maximum Likelihood (ML) estimationSkin detection resultsThis same procedure applies in more general circumstances• More than two classes• More than one dimensionGeneral classificationH. Schneiderman and T.KanadeExample: face detection• Here, X is an image region– dimension = # pixels – each face can be thoughtof as a point in a highdimensional spaceH. Schneiderman, T. Kanade. "A Statistical Method for 3D Object Detection Applied to Faces and Cars". IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2000) http://www-2.cs.cmu.edu/afs/cs.cmu.edu/user/hws/www/CVPR00.pdfLinear subspacesClassification can be expensive•