An efficient motion estimator with application to medical image registration.
ABSTRACT Image registration is a very important problem in computer vision and medical image processing. Numerous algorithms for registering single and multi-modal image data have been reported in these areas. Robustness as well as computational efficiency are prime factors of importance in image data registration. In this paper, a robust/reliable and efficient algorithm for estimating the transformation between two image data sets of a patient taken from the same modality over time is presented. Estimating the registration between two image data sets is formulated as a motion-estimation problem. We use a hierarchical optical flow motion model which allows for both global as well as local motion between the data sets. In this hierarchical motion model, we represent the flow field with a B-spline basis which implicitly incorporates smoothness constraints on the field. In computing the motion, we minimize the expectation of the squared differences energy function numerically via a modified Newton iteration scheme. The main idea in the modified Newton method is that we precompute the Hessian of the energy function at the optimum without explicitly knowing the optimum. This idea is used for both global and local motion estimation in the hierarchical motion model. We present examples of motion estimation on synthetic and real data (from a patient acquired during pre- and post-operative stages) and compare the performance of our algorithm with that of competing ones.
- European journal of cardio-thoracic surgery: official journal of the European Association for Cardio-thoracic Surgery 03/2011; 39(3):424; author reply 425. · 2.40 Impact Factor
Conference Paper: Optimal gradient pursuit for face alignment.[Show abstract] [Hide abstract]
ABSTRACT: Face alignment aims to fit a deformable landmark- based mesh to a facial image so that all facial features can be located accurately. In discriminative face alignment, an align- ment score function, which is treated as the appearance model, is learned such that moving along its gradient direction can improve the alignment. This paper proposes a new face model named "Optimal Gradient Pursuit Model", where the objective is to minimize the angle between the gradient direction and the vector pointing toward the ground-truth shape parameter. We formulate an iterative approach to solve this minimization problem. With extensive experiments in generic face alignment, we show that our model improves the alignment accuracy and speed compared to the state-of-the-art discriminative face alignment approach. I. INTRODUCTION Model-based image registration/alignment is a fundamen- tal topic in computer vision, where a model is deformed such that its distance to an image is minimized. In particular, face alignment is receiving considerable attention, because it not only enables various practical capabilities such as facial feature detection, pose rectification, face animation, etc, but also poses many scientific challenges due to facial appearance variations in pose, illumination, expression, and occlusions.Ninth IEEE International Conference on Automatic Face and Gesture Recognition (FG 2011), Santa Barbara, CA, USA, 21-25 March 2011; 01/2011
- [Show abstract] [Hide abstract]
ABSTRACT: Efficient and accurate fitting of active appearance models (AAM) is a key requirement for many applications. The most efficient fitting algorithm today is inverse compositional image alignment (ICIA). While ICIA is extremely fast, it is also known to have a small convergence radius. Convergence is especially bad when training and testing images differ strongly, as in multi-person AAMs. We describe ldquoforwardrdquo compositional image alignment in a consistent framework which also incorporates methods previously termed ldquoinverserdquo compositional, and use it to develop two novel fitting methods. The first method, compositional gradient descent (CoDe), is approximately four times slower than ICIA, while having a convergence radius which is even larger than that achievable by direct quasi-Newton descent. An intermediate convergence range with the same speed as ICIA is achieved by LinCoDe, the second new method. The success rate of the novel methods is 10 to 20 times higher than that of the original ICIA method.Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on; 07/2009
Univ ersityofFlorida?Gainesville?Fl ?????
Keywords?ImageRegistration?OpticalFlo w?SSD? B?splines? NewtonIteration?Leven?
This work was supp ortedin partbythegrantNIH?R???LM??????Submittedto thejournalof Medical
Imageregistration isav ery importantproblemincomputervisionandmedical im?
reported intheseareas?Robustness aswell ascomputationale?ciency areprime factors
of importance inimage dataregistration?Inthis paper? arobust ande?cient algorithm
forestimatingthe transformation b etween two imagedatasets is presented? Estimat?
?ow?eldwithaB?splinebasiswhichimplicitly incorporates smo othnessconstraintson
the?eld?Incomputing themotion?weminimize the expectation ofthesquareddi?er?
functionattheoptimum withoutexplicitly knowingtheoptimum?Thisideais used for
examplesofmotionestimationonsynthetic and realdataandcomparetheperformance
Imageregistrationisaverycommonandimp ortantproblemin diverse ?eldssuchas medical
imaging?computer vision?videocompression?art? entertainmentetc?The problem ofregis?
teringtwoimages?bethey ?D or ?D?is equivalenttoestimatingthemotionbet weenthem?
LaiandVem uri? ?????thatcouldpoten tiallybe appliedtotheproblem ofregistrationof
volume images?speci?cally ? MRbrain scans?whic histhe topicoffo cusin this thesis?We
drawup onthis largebodyofliteratureofmotionestimationtechniquesforproblemformula?
Pellizarietal??Pelizarrietal???????havedevelopeda method fortheregistrationof two
brains identi?ed fromtwodi?eren tdatasets byminimizingthedistancebet weenthesur?
sphericalshap es?Ev ansetal??Evans etlal??????? develop edaregistrationschemebasedon
approximatingthe?Dw arpb etween themo delandtargetimagebya ?Dthinplatespline
?ttedtolandmarks? Szeliski etal? ?Szeliski andCoughlan? ?????dev elop edafastmatc hing
algorithmforregistrationof ?Danatomicalsurfacesfound in?D medicalimagedata?They
torepresen t di?erentregistrationtransformations?namelyrigid?a?ne?trilinear?quadratic
putation ofdistance bet weensurfaces?Davatzikos etal??Davatzikosand Prince? ?????
inthe brainandin thesecondstage?anelasticdeformationtransformationthatdetermines
Feldmaretal?? ?F eldmarandAy ache??????developeda novelsurfacetosurfacenonrigid
registrationscheme?usinglocallya?netransformation?Maintzetal?? ?Main tzetal???com?
parededge?Borgefors??????andridge?based?Gueziec and Ayache??????Mongaet al???
v andenElsen??????registrationofCT andMR brainimages?Theydescribeanovelmethod
features?surface?contoursintheimages andhencetheaccuracyofregistration isdictatedby
theaccuracy of thefeaturedetector? Also? additional computationaltimeisrequiredfor
detecting these featuresprior toapplicationoftheactualregistration scheme?In this paper?
we prop osearegistr ationmethodthatis robustandfastandisapplicabledirectlytotheraw
lan???????Amore generalsc hemethan window?basedcorrelationapproach isthe optical
?owformulation? inwhichtheproblemof registeringtwoimagesis treatedasequivalentto
optical?ow fromapair ofimages?HornandSchunk??????Barronetal???????Duncanand
Chou??????Lai andV emuri??????Szeliski andCoughlan? ????? BlackandAnandan? ?????
Gupta andPrince? ??????
portedinvioloaandWells?ViolaandWells? ?????andWellsetal???WellsIII etal????????
informationbetweenthemo deland the imagethataretoberegisteredismaximizedusing
astochasticanalogofthegradien tdescentmethodin?Wells IIIetal???????andotheropti?
mizationmethodssuchasthePow ellsmethodin ?Collignon etal??????? andamultiresolution
sc hemein ?Studholmeetal????Reportedregistration experimen ts intheseworksarequite
impressiveforthecaseof rigidmotion?Theresearchrep orted onthem utualinformation?
basedmethod requires thatthemotionbeexpressedinaparameterizedformandbeglobal?
Thereisnoprovisionforhandling local?nonparametric? andmultiplemotions?
Ba jscyandKov acic?Ba jscyand Kovacic? ????? studiedregistration undernonrigid de?
formationsusingvolumetricdeformations basedonelasticitytheory ofsolids?Otherdirect
etal??Christensenel al????????Inthisapproachtheregistrationtransformationis modeled
byaviscous?uid ?ow mo delexpressedasapartial di?erential equation?PDE??The model
primarilydescrib es theregistration asa lo cal ?uiddeformationexpressedbya nonlinearPDE
and more recentlybya linearversion ?Bro?NielsenandGramko w???????Thirion?Thirion?
????? intro duced aninterestingdemon?based registrationthatcanbeview edasb eingsimilar
forregistrationof?Dand?Ddatasets arevery largefor implementationonuniprocessor
w orkstations?The fastest implementationoftheviscous?uid ?o wmo delby Schormann et
al? ?Schormann etal??????? usingamulti?grid schemetakes???mins?ofCPUtime ona
Insummary?theexistingregistration algorithms eithercan?tbe applied tora w im?
age?v olumedatadirectly? orcan?t handlelocaldeformation andglobal transformationat
the same time?orrequireextensivecomputationandhence? areunsuitableformost ofthe
applications?In thispaper? a robustand fastalgorithmisproposed?whichincorporatesa
Modi?ed Newton iterativemethodintoaspline?basedoptical?owframeworkandachieves
Inourapproach?weusethe hierarc hical motionmo del?ah ybridoflocalandglobal?ow?
mo delconsistsofgloballyparameterized motion?ow models at oneendofthe?spectrum?and
alocalmotion?owmodelattheother end?Theglobal motionmo del isde?nedby associating
a singleglobal motiontransformation withineach patch ofa recursivelysubdivided input
image?whereas? inthelo cal?o wmo del?the?ow?eldisnotparameterized?The?ow?eld
di?erences error term?i?e??E
?ow ?eldmaybeestimatedat eachpixel?vo xel vianumericaliterativeminimizationofE
erarchical?ow?eldcomputationmodelbasedontheideaof precomputing theHessianat
theoptimumpriorto kno wingtheoptim um??b? dev elopment ofanov el? fast androbust
numericalsolution techniqueusing a Modi?ed Newtoniteration?wherein the computationof
theHessianmatrix and gradients useaspline?basedrepresentationofthe?u?v? ?eld??c?a
themotion?registration? InSection ??wepresent severalexp erimen talresultson syn thesized
motion applied toreal brain MRI dataasw ellas resultsof registeringMR brainscansof the
same individual?Section ?con tains theconclusions?