ieee transactions on education, vol. 45, no. 4, … · matlab and labview are used as the...

IEEE TRANSACTIONSON EDUCATION, VOL. 45, NO. 4, NOVEMBER 2002 1

TheSIVA DemonstrationGalleryfor Signal,Image,andVideoProcessingEducation

UmeshRajashekar, Student Member, IEEE, GeorgeC.Panayi,FrankP. Baumgartner,andAlan C.Bovik, Fellow, IEEE

Abstract—The techniques of digital signal processing (DSP) anddigital image processing (DIP) have found a myriad of applicationsin diverse fields of scientific, commercial, and technical endeavor.DSP and DIP education needs to cater to a wide spectrum ofpeople from different educational backgrounds. This paperdescribes tools and techniques that facilitate a gentle introductionto fascinating concepts in signal and image processing. NovelLabVIEW- and MATLAB-based demonstrations are presented,which, when supplemented with Web-based class lectures, helpto illustrate the power and beauty of signal and image-processingalgorithms. Equipped with informative visualizations and auser-friendly interface, these modules are currently being usedeffectively in a classroom environment for teaching DSP and DIPat the University of Texas at Austin (UT-Austin). Most demon-strations use audio and image signals to give students a flavor ofreal-world applications of signal and image processing. This paperis also intended to provide a library of more than 50 visualizationmodules that accentuate the intuitive aspects of DSP algorithmsas a free didactic tool to the broad signal and image-processingcommunity.

Index Terms—Demonstration library, interactive education,multidisciplinary, signal and image-processing education, visual-ization.

I. INTRODUCTION

THE PRINCIPLESof digital signalprocessing(DSP)anddigital imageprocessing(DIP) have spreadtheir rootsfar

andwide,asevidencedby theirapplicationstoaverywidespec-trum of problems.Astronomy, genetics,remotesensing,videocommunications,and ultrasonicimaging are just a tiny sam-pling of applicationsthatreflectthemultidisciplinarynatureofDSPandDIP. Applicationsin DSPandDIP combineconceptsfrom avarietyof areas,suchasvisualpsychophysics,audioandacoustics,optics,andcomputerscience.Although well rootedin advancedmathematicsunfamiliarto amajorityof thegeneralaudiencethat usesit, the theoryof DSPandDIP needsto bemade“accessible” to practitionersfrom diversebackgrounds.Presentingsuchaninterdisciplinarytopic with perspicuityto aheterogeneousaudienceis challenging.With therecenttrendofDSPdrifting lowerdown thecurriculumto evenhighschool,asin theInfinity project[3], andtheimportanceof designingDSPasa first coursein electricalandcomputerengineering[4], theneedfor toolsandtechniquesto presentsignalprocessingwith

Manuscriptreceived October19, 2001;revisedMarch 1, 2002.Portionsofthis manuscriptappearedin [1], [2].

The authorsare with the Departmentof Electrical and ComputerEngi-neering,University of Texas at Austin, Austin, 78712-1084,USA (e-mail:[email protected];[email protected];[email protected]).

PublisherItem Identifier 10.1109/TE.2002.804392.

minimal mathto a nontechnicalaudienceis becomingincreas-ingly significant.

Themainhurdlefacedby a novice in DSPis thatthemathe-maticsthatdescribesfundamentalconceptscancloudintuition.To reinforceconcretefundamentalconcepts,mostintroductorycoursesassigncomputer-basedDSPexercises.However, moreoften thannot, the learningcurve involved in becomingfamil-iarizedwith thesoftwaredetractsthestudentfrom assimilatingtheconcept.More effective techniquesto uncover theintuitionbehind“murky” equationsarevisualizationtoolsthat facilitatetheauralandvisualconsumptionof information.A ready-to-usesetof demonstrationsillustratingtheconceptsthattheinstructordeemsimportantcanhelp the studentto begin experimentingand assimilatingimmediatelywithout having to botheraboutprogrammingintricacies.This situation also encouragesstu-dentsto experimentwith their desiredinputs at their leisure.Further, suchtools bolsterthe successof distancelearningbyfacilitatinginteractiveeducationandarethetopicof discussionin this paper.

Therehave beenmany significantcontributionsto DSPedu-cationaltoolsdevelopedprimarily with MATLAB [1], [5]–[8],LabVIEW [2], [9], Java[10]–[12], andMathematica[13]. Java-basedtools enjoy the advantageof inherentplatform indepen-dence[14] over mostother implementations.Further, they donotneedtheuserto haveany specific softwareinstalledontheirlocalmachine,makingthisoptionmosteconomicalto students.Java-basedtools are also inherentlysuitablefor a Web-basededucationsystem,sincethey canbeeasilyintegratedinto Webbrowsers.Commongateway interfacehasalso beenusedforhandlingsignal-processingroutines,with Javausedfor theuserinterface[15]. While thisapproachsoundsoptimistic,onemustbearin mind that developingeducationaltools with softwarespeciallydesignedfor signal-processingapplicationsis obvi-ously lesstedious.Many of thesedemonstrationshave focusedsignificantly on conceptssuchasthe transform[16] [17] anda few otherelementaryconceptsin DSP[6].

ThispaperdescribestheSignal,Image,andVideoAudiovisu-alizationDemonstrationGallery(SIVA), comprisedof twopow-erful visualizationmodulesrich in fundamentalconcepts,devel-opedandusedat theUniversityof TexasatAustin (UT-Austin)for signal-andimage-processingcourses.Themodulesconsistof aLabVIEW-baseddemonstrationsuitefor anundergraduatecoursetitled “Digital ImageProcessingandVideoProcessing”andaMATLAB demonstrationpackagefor agraduatecourseti-tled“Digital SignalProcessing.”SIVA is tailoredfor anin-classandonlineinstructionambiencewith apowerfulpoint-and-clicktype of graphicaluserinterface(GUI). The demoshave beenseamlesslyintegratedinto theclassnotesto provide contextual

0018-9359/02$17.00© 2002IEEE

2 IEEE TRANSACTIONSON EDUCATION, VOL. 45, NO. 4, NOVEMBER 2002

(a)

(b)

Fig. 1. Typical GUI developmentenvironmentin LabVIEW. (a) Frontpanel.(b) Block diagram.

illustrations.Thissuite,with morethan50demosspansagamutof conceptsin signalandimageprocessing,muchwiderin scopethanmostof thecurrentendeavorsof similar flavor mentionedpreviously.

MATLAB andLabVIEW areusedasthedevelopmenttoolsfor thesedemonstrationmodules.MATLAB hasbecomethesoftwareof choicefor DSPsimulationsasreflectedby theubiq-uity of thissoftwarein universitylabs.Inexpensivestudentver-sionsof MATLAB arealsoavailable,makingit oneof themostaccessiblesoftwareplatformsfor DSPdevelopment.Theenvi-ronmentis intuitive to work with andprovidesspecializedtool-boxesfor signalandimageprocessing,amongstmany others.MATLAB alsohastheadvantagethatit is highly optimizedforvectorizedcode,makingit suitablefor DSPalgorithms.

Similarly, LabVIEW is becominga very popularsoftwarein most universities.One of the reasonsfor its popularity isthatbuilding front-endGUIs is extremelysimple.Becausetheprogrammingis graphical,it doesnotoverwhelmprogrammers

with syntacticaldetails.LabVIEW alsoprovidesIMAQ Vision,a suite of image-processingroutines (besidesmany others),which made it an attractive choice for building image-pro-cessingdemos.Building educationaltoolswith MATLAB andLabVIEW hasthe advantagethat mostendusersare familiarwith thesetools and can build on the code or make minormodificationsandcustomizeit for their intendedaudience.

To obviatetheneedfor acopy of MATLAB at theclientend,MATLAB hasdevelopedthe“MATLAB Server” solution[18].Similarly, National Instrumentshas releasedtheir LabVIEWPlayer, whichagainmakesunnecessarytheneedfor alocalcopyof LabVIEW on theclient end.Theability to programsimply,impressiveintegratedgraphicalfunctions,availability for awidevarietyof platforms,andextensibilityto aWeb-basededucationsystem,suchasthatusedattheUT-Austin,madeMATLAB andLabVIEW obviouschoicesfor thedevelopingplatform.

Therestof thispaperisorganizedasfollows.Thebasiccoursestructureof the two coursesfor which thesetools have been

RAJASHEKAR et al.: THE SIVA DEMONSTRATION GALLERY FORSIGNAL, IMAGE,AND VIDEO PROCESSINGEDUCATION 3

(a)

(b) (c)

Fig. 2. Effectsof gray-level quantization.(a) Frontpanel.(b) “Barbara”at8 b/pixel. (c) “Barbara” at 2 b/pixel.

(a)

(b) (c)

Fig. 3. Binary imagemorphology. (a) Front panel.(b) Original “cell.” (c)“Cell” eroded.

developedfor in-classaudio-visualdemonstrationsis described

in SectionII. SectionIII is anoverview of the DIP featuresofSIVA; SectionIV is an overview of the DSPdemosin SIVA;andSectionV presentstheconclusion.

II. OVERVIEW OF COURSES

Thoughintendedfor anelectricalandcomputerengineeringcurriculum,the junior/seniorcourse“Digital ImageandVideoProcessing”(EE371R)andthegraduatecourse“Digital SignalProcessing” (EE 381K-8) attractsstudentsfrom variousothermajors,suchaspsychology, aerospace,geology, astronomy, andcomputerscience.There is also a steadyenrollmentof pro-fessionalsfrom the local industriesat Austin, TX. The objec-tive of both coursesis “ to make signal processing accessibleto an audience with heterogenous backgrounds by augmentingtheory with numerous audio-visual examples.” To encourageaWeb-basededucationalsystem,WebCT[19] is usedasan in-structionmediumto make all coursematerialanddemonstra-tionsavailableover theWorld Wide Web.

A. EE 371R: DIP and Video Processing

Introductory material covered in this junior/senior courseon digital imageandvideo processingincludesbinary imageprocessing,image analysis,and image enhancement,whilethe more advanced material covers such topics as Houghtransforms,edgedetection,andvideoprocessing.

Along with coursematerial [20] designedassiduouslybythecourseinstructorto keepthelevel of mathaccessibleto theaudience,a recenthandbook[21] was introducedfor the firsttime in fall 2000 to provide the audiencewith an invaluablereferenceand classroomtext for the course.To assistvisualinterpretationof ideasdiscussed,in-classdemonstrationsusingSIVA are usedto complementlectures.Simple but intuitiveMATLAB-based assignmentsare designedto reinforce theconceptswithoutoverwhelmingstudentswith programmingin-tricacies.Studentsarealsoencouragedto investigateandapplytheconceptslearnedin thecourseto their respective fields,anda monetaryaward is given as an addedincentive to the bestclassproject. Studentsare provided with digital camcorders(CanonZR-10) and firewire cardsto download digital videoonto computers,video-editingsoftware, Web cameras(VistaImaging) and state-of-the-artvideoconferencingequipment(CanonVC-C4) to develop their projects.The availability ofequipmenthasgreatlyassistedin motivating studentsto buildintellectuallystimulatingprojects.

B. EE 381K-8: DSP

This graduatecoursein digital signalprocessingis offeredevery springat UT-Austin. In contrastto EE 371R,this courseis mathematicallyrigorous.The topicscoveredincludesignalrepresentations,Fourier seriesexpansions, transforms,filterdesign,nonlinearfilters,discrete-timerandomprocesses,quan-tizationeffects,multirateprocessing,andsubbandfilter banks.Coursenotes[22], meticulouslydesignedfor theaudience,areprovided online on the courseWeb site. To emphasizethe ef-fectsof DSPalgorithmson real-world signals,theDSPsectionof SIVA wasdesignedin harmony with themoduleshandledin


(a)

(b) (c) (d)

Fig. 4. Histogramshaping.(a) Frontpanel.(b) Original “books.” (c) Histogram—flat. (d) Histogram—invertedGaussian.

class.Many audiosignalsareusedduringthelecturesto inter-prettheory. All demosarelinkedto thecoursenotesfor in-classdemonstrations.

III. SIVA IMAGE PROCESSINGDEMONSTRATION GALLERY

LabVIEW [23] is agraphicalprogramminglanguageusedasa powerful andflexible instrumentationandanalysissoftwaresystemin industryandacademia.LabVIEW usesa graphicalprogramminglanguage,G, to createprogramscalledvirtual in-struments(VIs) in apictorialform,eliminatingmuchof thesyn-tacticaldetailsof othertext-basedprogramminglanguages,suchasC andMATLAB. LabVIEW includesmany toolsfor dataac-quisition, analysis,anddisplayof results.LabVIEW is avail-able for all the major platformsand is easily portableacrossplatforms.LabVIEW hasthe ability to createstand-aloneex-ecutableapplications,which runat compiledexecutionspeeds.

Anotheradvantageof LabVIEW is that it includesbuilt-inapplications,suchas the IMAQ Vision for imageprocessing.IMAQ Vision includesmore than 400 imaging functionsandinteractive imaging windows and utilities for displayingandbuilding imagingsystems,giving designersthe opportunitytocreateexamplesfor many important conceptsin image pro-cessingandusingthemfor educationalpurposes.An excellentintroductionto LabVIEW is provided in [24] and [25]. Mosttechnicalinformationfor thedevelopmentof theVIs andIMAQareprovidedin [26] and[27], respectively. Informationspecificto theVIs developedfor this coursecanbeobtainedin [28].

A. LabVIEWDevelopmentEnvironment

EachVI containsthreeparts.

• Thefront panelcontainstheuserinterfacecontrolinputs,suchasknobs,sliders,andpushbuttons,andoutputindi-catorsto producesuchitemsaschartsandgraphs.Inputscanbefedinto thesystemusingthemouseor keyboard.Atypical front panel,usedto provide intuitiveGUIs to varytheparametersof thealgorithm,is shown in Fig. 1(a).

• Theblockdiagramshown in Fig. 1(b) is theequivalentofa“sourcecode”for theVI. Theblocksareinterconnected,usingwires to indicatethedataflow. Front-panelindica-torspassdatafrom theuserto their correspondingtermi-nalsontheblockdiagram.Theresultsof theoperationarethenpassedbackto thefront-panelindicators.

• Sub-VIsareanalogousto subroutinesin conventionalpro-gramminglanguages.

B. Examplesof Demos

SIVA encompassesa wide rangeof VIs that canbe usedinconjunctionwith classlecturesonimageprocessing.Becauseofspaceconstraints,onlyasmallnumberof theLabVIEWVIs thatweredevelopedaredescribed.Thereaderis invited to exploreanddownloadtheotherDIP demosin SIVA from theWebsitementionedin [29].

1) A–D Conversion: Samplingandquantizationarefunda-mentaloperationsin signalandimageprocessingthattransform


(a)

(b) (c) (d)

Fig. 5. DiscreteFouriertransform.(a) Frontpanel.(b) Original “cameraman.”(c) DFT magnitude.(d) DFT phase.

acontinuoussignaltoadiscreteone.Thoughthetheoryismath-ematicallycumbersome,the effectsof samplingandquantiza-tion canbevisualizedeffectively usingtheVIs shown in Fig. 2.Falsecontouringeffectsresultingfrom gray-level quantizationarevisually obvious in Fig. 2(c). This VI readsin an8-b/pixelimageandcreatesanoutputwhosebit depth(1,2,4,or8b/pixel)can be specified from the front panel.The samplingVI (notshown) canbeusedto visualizealiasingcausedby sampling.

2) Binary Image Processing: Binary imageshave only twopossible“gray levels” andarethereforerepresentedusingonly1 b/pixel. BesidessimpleVIs usedfor thresholdinggray-scaleimagesto binary, other VIs were developedto demonstratethe effects of binary morphology. Morphological operationsare defined by moving a structuringelementover the imageand performinga logical operationon the pixels coveredbythestructuringelement,resultingin anotherbinary image.ThemorphologyVI shown in Fig. 3 canbeusedto demonstratetheeffectsof variousmorphologicaloperationson binary images,e.g., median,dilation, erosion,open,close, open–close,andclose–open.The useralso hasthe option of varying the sizeof thestructuringelement,which canbeany of the following:row, column,square,cross,or X-shape.

3) Histogram and Point Operations (Gray-Scale): Todemonstrateeffectsof elementarygray-scaleimageprocessing,VIs thatperformlinear (offset,scaling,andfull-scalecontraststretch)and nonlinear(logarithmic rangecompression)pointoperationson imageswere developed.A more advancedVI,shown in Fig. 4, demonstratestheeffectsof histogramshaping.The histogramsof the input image and the resulting image

after the linear point operationarealsodisplayedon the frontpanelto verify theresultsof theshaping(e.g.,thehistograminFig. 4(d) is inverseGaussian-like).

4) Image Analysis (Frequency Interpretations): This con-sistsof discreteFouriertransform(DFT) anddirectionalDFTs.

a) DFT: Though a difficult conceptfor many studentsto comprehend,a lucid understandingof Fourier transformsis critical to the more advancedtopicsof imagefiltering andspectral theory. The study, therefore,begins by introducingthe conceptof digital frequency, usingtwo-dimensional(2-D)digital sinusoidalgratings.TheDFT demonstrationVI (shownin Fig.5) computesanddisplaysthemagnitudeandthephaseoftheDFT for gray-level images.TheDFT canbedisplayedwithits low frequenciesclusteredtogetherat thecenterof theimageor distributedat theperiphery. An optionis providedto displaythe logarithmically compressedfull-scale contrast-stretchedversion of the magnitude spectrum to reveal low-contrastvalues.

b) Directional DFTs: When the DFT of an image isbrighteralonga specific orientation,it implies that the imagecontainshighly orientedcomponentsin thatdirection.Orientedbinary imagescanbe usedto maskthe DFT of theseimages,which, whenoperatedon by the inverseDFT, produceimageswith only highly orientedfrequenciesremaining.To demon-stratethedirectionalityof theDFT, theVI shown in Fig. 6 wasimplemented.As shown on the front panel in Fig. 6(a), theinput parameters,Theta1 andTheta2, areusedto control theangleof thewedge-like,zero-onemask.It is instructive to notethat zeroingout someof the orientedcomponentsin the DFT


(a)

(b) (c)

(d) (e)

Fig. 6. Directional DFT. (a) Front panel. (b) Original “Escher.” (c) DFTmagnitude.(d) MaskedDFT. (e) InverseDFT aftermasking.

results in the disappearanceof the conduitsin the “Escher”imagein Fig. 6(e).

5) Image Filtering: SIVA includesmany demosto illustratethe useof linear andnonlinearfilters for imageenhancementand image restoration.The use of low-passfilters for noisesmoothingand inverseand pseudoinverse filters for decon-volving imagesthat have beenblurred areexamplesof somedemos for linear image enhancementin SIVA. SIVA alsoincludesdemosto illustrate the prowessof nonlinear filtersover their linear counterparts.Fig. 7 demonstratesthe resultof filtering, with both a linear filter (average)anda nonlinear(median)filter, a noisy imagecorruptedwith “salt & peppernoise.”

6) Image Compression: TheVI on block truncationcoding(BTC) shown in Fig. 8 providesanintroductionto lossyimagecompression.The usercan selectthe numberof bits usedtorepresentboth themeanof eachblock in BTC (in ) andtheblockvariance(in ). Thecompressionratio is computedanddisplayedon thefront panelin the indicatorin Fig. 8(a).

OtheradvancedVIs developedincludemany linearandnon-linearfiltering for imageenhancement,otherlossyandlosslessimagecompressionschemes,a largenumberof edgedetectorsfor imagefeatureanalysis,andHoughtransforms.Theelemen-tary oneshave beenpresentedherefor the purposeof illustra-tion.

IV. SIVA SIGNAL PROCESSINGDEMONSTRATION GALLERY

MATLAB [30] is ahigh-performancelanguagefor technicalcomputing.It integratescomputation,visualization,and pro-gramminginto aneasy-to-useenvironment[31]. Thebasicdataelementin MATLAB is anarray. Many matrix-basedfunctions,suchasmatrix multiplicationandarraydot product,canbeex-ecutedin a fractionof thetime it would take to write a similarprogramin a scalar, noninteractive languagesuchasC or For-tran. MATLAB featuresa family of application-specific solu-tionscalledtoolboxesfor suchapplicationsassignalprocessing,neuralnetworks,andwavelets.Thesetoolboxesarea library offunctionswritten asM files.

Thesignal-processingtoolbox[32], for example,includesaninteractiveenvironmentfor analyzingandmanipulatingsignalsand designingfilters. MATLAB hasa numberof easy-to-useplotting andgraphicalfunctions,which make MATLAB anat-tractive choicefor developingattractive visualizationapplica-tions.Thevectorizednatureof MATLAB andtheabundantcol-lection of functionsandvisualizationoptionsmake it a goodchoicefor visualizingDSPconcepts.MATLAB providesapow-erful GUI developmenttool critical to the creationof educa-tional toolsfor classroominstructionneeds.CurrentversionsofMATLAB alsoprovidepowerful user-friendlydebuggingtools.

A. GUI Development Using MATLAB

MATLAB providesthe GUI DesignEnvironment(GUIDE)to develop impressive GUIs quickly using drag-and-dropob-jects,suchasbuttons,sliders,andpop-down menus[33]. Fig. 9shows a typical GUI developmentenvironment.The GUIDEcontrolpanelprovidesthedrag-and-dropobjects,thepropertiesof whicharecontrolledby theGUIDEpropertyeditor. An M fileperformingaparticulartaskfor eachobjectin theGUI iswrittenseparately, usingmany of thein-built MATLAB functions.TheGUIDE callbackeditormanagestheactionsassociatedwith theselectionof aparticularobject(e.g.,clickingabutton)by linkinganobjectto its respective M file (asshown by thesequenceofarrows in Fig. 9). OncetheGUI is developed,onecanusethemouseandon-screenoptionsto visualize,hear, andmanipulatesignals(audioand images).A numberof signalscanbe ana-lyzed,e.g.,in audio,maleandfemalevoices,music,andstan-dardwaves(e.g.,sine,chirp,square,andtriangle).A few demosalsouseimagesasinputsto illustratemultidimensionalDSPal-gorithms.For classroominstruction,thedemosarehyperlinkedfrom anHTML document.

B. Examples of Demos

Using the powerful graphicsand simple functionality ofMATLAB, a numberof DSPdemonstrationsthat canbe usedfor a classroomteachingenvironment have beendevelopedto demonstratevisually the fundamentalDSP conceptsusing


(a)

(b) (c)

(d) (e)

Fig. 7. Nonlinearfiltering. (a) Frontpanel.(b) Original “Dhivya.” (c) Salt&peppernoise.(d) Averagefiltering. (e) Medianfiltering.

mainly audioanda few syntheticsignals.Currently, all demosrun on the Windows platform but can be easily transportedonto otherplatformsaswell. In the following subsections,anoverview of a few modulesthat weredevelopedis presented.The readeris invited to experimentwith the othersby down-loadingthemfrom theWebsitementionedin [29].

1) SignalRepresentations:In this introductorymodule,thefocusis ontherepresentationsof aninputsignalin differentdo-mains:time, frequency, andtime–frequency domains.Theusercan selectan audio sampleto be viewed in any of thesedo-mains,changeits samplingfrequency and the numberof bitsper sample,andhearthe original andthe modified signal.Se-lectingachirpsignalandsamplingit appropriatelycanbeusedeffectively to visualizeandhearaliasing,asshown in Fig. 10.Othersimpledemoshave beendevelopedto illustratethe dif-ferencebetweencontinuous,discretetime,anddigital signals.

2) Fourier Series: One of the mathematicallyintriguingtechniquesof signaldecompositionis thatof theFourierseries.The conceptof many sinusoidsperfectlyaligning to interfereconstructively and destructively to create the time-domainsignal is illustratedpowerfully in the demoshown in Fig. 11,using a saw-tooth waveform as an example.The dotted lineshows the reconstructionusingsix Fourier seriescoefficients.The slider “coefficients used” in the GUI enablesthe usertoaddmoresinusoidsanddemonstratetheconstructionusingthesinusoidalbasis.

3) Transform: The transformdemo(shown in Fig. 12)demonstratesthe powerful graphicscapability of MATLAB.The effect of zerosand poleson the impulseand frequencyresponseof a systemcanbedemonstrated.Theusercanplacepolesandzeroson the planeusingtheclick of a mouse.Thethree-dimensional3-D effectof thepolesandzeros[Fig. 12(b)],along with other options,suchas frequency [Fig. 12(c)] andimpulseresponses[Fig. 12(d)],canbevisualized.

4) Gibb’sPhenomenon:Whengiventhefrequency-domainspecifications,windowing the inverseFourier transformis acommontechniqueusedto designfilters in DSP. The GUI inFig.13depictstherippleeffectsof Gibb’sphenomenon,causedby truncating.

5) Filter Design: Filter designdemosfor finite-impulsere-sponse(FIR) filters andinfinite-impulseresponse(IIR) filterscanbe usedto illustratethe effectsof filtering real-world sig-nals.Thedemoshaveoptionsto view thepole-zeroplotsof fil-ters, the magnitudeandphaseresponseof the designedfilter[Fig. 14(b)],andto selectaudiosignalsandseeandheartheef-fectsof filtering them.A frequency samplingtechniquefor FIRfilter designis shown in Fig. 14.Theeffect of filtering a linearchirp signalusingthedesignedfilter is illustratedin Fig. 14(c).Also includedin the GUI areoptionsto control the filter typeandcutoff frequencies.

6) Decimation: The differencebetweensubsamplinganddecimationis describedintuitively in Fig. 15. The aliasingintroduced becauseof subsamplingis visually obvious inFig. 15(c), where the spectrumreplicasoutsidethedigital frequency bandhave enteredthe band (notethe reverseddirection of the triangularbands).Selectingthelow-passfilter optionin theGUI filters theoriginal signalwithan appropriatelow-passfilter beforesubsamplingin order topreventany aliasing,evidentin Fig. 15(d).Thealiasingeffectsareaudiblewhenthesubsampledsignalsareplayed.

Besidesthe elementaryexamplesdiscussedpreviously, anumber of other demosexplain conceptsin Fourier series,discrete Fourier transforms, filter design, multirate DSP,short-timeFourier transforms,etc. A medianfilter demo(foraudioandimages)wasalsodevelopedto illustratethesuperiorperformanceof nonlinearfilters over linearsystems.

V. CONCLUSION

The importanceof stressingpracticalapplicationsin a DSPcoursehasbeenprevalentfor a longtimenow. MakingDSPac-cessibleto anever-growing nonexpertaudienceis highly desir-able.In thispaper, alibraryof demonstrationsbuilt, andsuccess-fully usedat UT-Austin, intuitively introducethe conceptsof


(a)

(b) (c)

(d) (e)

Fig. 8. Block truncationcoding.(a)Frontpanel.(b) Original “Peppers.”(c) 2b for mean;0 b for variance.(d) 3 b for mean;4 b for variance.(e)6 b for mean;5 b for variance.

digital signalprocessinganddigital imageandvideoprocessingto a disparateaudienceconsistingof graduateandundergrad-uatestudentsfrom a varietyof backgrounds.Theseclassesareamazinglypopular, typically attractingmorethan80 studentsperclassin recentsemesters.Theinertiato developsuchdemosis understandablesincea high investmentof time andeffort is

required.Therefore,theSIVA visualizationpackageis providedfreeontheWorld WideWeb[29] for pedagogicpurposes.Morethan40usersfrom variouscountriesarealreadyusingSIVA fortheir projectsandasa teachingtool for signalandimagepro-cessingcourses.It is thehopeof theauthorsthatthispaperwillsucceedin attractingthe attentionof many moreDSPinstruc-tors to theavailability of SIVA.

Encouragedby thesuccessof thesevisualizationsin tandemwith classlectures,work is in progressto build anothervisual-ization in the areaof digital video processing.The video pro-cessingdemosin SIVA will usetheLabVIEW environmentandIMAQ Vision to demonstratesuchkey conceptsasmotionesti-mationandcompensation,motion-compensatedfiltering, videocompressionandreconstruction,the effectsof noise,andpre-processing.Theresultsfrom thesedemoscanbecomparedbothobjectively (e.g.,peaksignalto noiseratio)andsubjectively forassessmentof videoreconstructionquality.Within many demos,userswill have theoptionto vary computationmethodsfor ad-ditionalcomparison.Fig.16,for example,illustratestheestima-tion of opticalflow betweentwo scenesof a video.

To illustrate a practical optical flow estimatingtechnique,the motion-estimationdemoshown in Fig. 17 allows the useof differentsearchmethods,suchasthe three-stepsearch,thecrosssearchor bruteforce,anddifferentmatchmethods,suchastheminimummean-squareerror, maximumabsolutediffer-ence,maximummatching-pixel count,or maximumcorrelation[34]. In addition,many demoswill operateon live video ac-quiredin-classinto LabVIEW from a digital camcorderandalaptopIEEE1394/firewireport.Thesedemoswill addflexibilityand intrigue to the learningprocessandhelp emphasizeboththe simplicitiesandthe intricaciesof video processing,whichmaybeoverlookedwhenusinga predefinedsetof videosthathavecontrolledacquisitionenvironments.Thenetresultwill bea packageof simple,qualitative,yet practicalvideoprocessinglearningtools.

TheLabVIEW demoswill beconvertedto LabVIEW playerformat in the near future, making unnecessarythe needforlocal copiesof LabVIEW on theclient end.It is alsoproposedto expandsuchtools to encompasssignal-processingaspectsof waveletsin the comingmonths.Theavailability of the freeMATLAB-basedWavelabpackage[35] is a strongmotivationfor building thesedemosusing MATLAB. The authorsareoptimistic that theselibraries will be instrumentalin addingvalueto theway digital videoprocessingandwaveletswill betaughtat UT-Austin and,hopefully, at many otheruniversitiesaroundtheworld.

RAJASHEKAR et al.: THE SIVA DEMONSTRATION GALLERY FOR SIGNAL, IMAGE, AND VIDEO PROCESSING EDUCATION 9

Fig. 9. Typical GUI development environment in MATLAB.

Fig. 10. Signal representations.


Fig. 11. Fourier series.

(a) (b)

(c) (d)

Fig. 12. � transform.(a) � transformGUI. (b) 3–D pole-zeroresponse.(c) Magnituderesponse.(d) Impulseresponse.


(a)

(b) (c) (d)

Fig. 13. Gibb’s phenomenon.(a) GUI for Gibb’sphenomenon.(b) Impulseresponseof desiredlow-passfilter. (c) Truncatedimpulseresponse.(d) Desiredanddesignedresponses.

(a)

(b) (c)

Fig. 14. FIR filter designby frequency sampling.(a)Frequency samplingGUI. (b) Frequency responseof desiredanddesignedfilters.(c) Resultsof filtering ofchirp signal.

12 IEEE TRANSACTIONS ON EDUCATION, VOL. 45, NO. 4, NOVEMBER 2002

(a)

(b) (c) (d)

Fig. 15. Decimation. (a) Decimation GUI. (b) Frequency spectrum of audio signal. (c) Spectrum after subsampling by factor of 2. (d) Spectrum after decimationby factor of 2.

(a) (b) (c)

Fig. 16. Optical flow. (a) Front panel. (b) Successive video frames. (c) Estimated optical flow.


(a)

(b) (c) (d)

Fig. 17. Block-basedmotion estimation.(a) Front panel.(b) Original “Lena.” (c) “Lena” shiftednorthwestby a few pixels. (d) Arrows indicatingestimatedmotion for eachblock.

ACKNOWLEDGMENT

Thiscoursewaredevelopmentwork waslargelysupportedbygrantsfrom theTexasTelecommunicationEngineeringConsor-tium (TxTEC)andby softwareassistancefrom NationalInstru-ments,Inc.

REFERENCES

[1] U. Rajashekarand A. C. Bovik. Interactive DSP educationusingMATLAB demo.presentedat 1st SignalProcessingEducationWork-shop.[Online]. Available:http://spib.ece.rice.edu/DSP2000/

[2] G. C. Panayi,U. Rajashekar,andA. C. Bovik. Imageprocessingforeveryone.presentedat1stSignalProcessingEducationWorkshop.[On-line]. Available:http://spib.ece.rice.edu/DSP2000/

[3] The Infinity Project [Online]. Available: http://www.infinity-project.org/home.html

[4] T. Barnwell and B. Evans. DSP as a first course. presentedat1st Signal ProcessingEducation Workshop. [Online]. Available:http://spib.ece.rice.edu/DSP2000/program.html#dspcourse

[5] G.C.OrsakandD. M. Etter, “CollaborativeDSPeducationusingtheIn-ternetandMATLAB,” IEEE Signal Processing Mag., vol. 12,pp.23–32,Nov. 1995.

[6] R.RadkeandS.Kulkarni.An integratedMATLAB suitefor introductoryDSPeducation.presentedat1stSignalProcessingEducationWorkshop.[Online]. Available:http://spib.ece.rice.edu/DSP2000/

[7] M. W. J. Williams and G. C. Orsak.Perunaand pony express:TwoMATLAB-basededucationalsoftware packagesfor signal processingand communications.presentedat 1st Signal ProcessingEducationWorkshop.[Online]. Available:http://spib.ece.rice.edu/DSP2000/

[8] J. RosenthalandJ. H. McClellan,“AnimationsandGUIs for introduc-tory engineeringcourses,”in Proc. Int. Conf. on Electrical Engineering,Aug. 2001,pp. 6E411–6E416.

[9] NationalInstrumentsLabVIEW PlayerVI Gallery[Online]. Available:http://zone.ni.com/devzone/explprog.nsf/webLabVIEWenabled

[10] Y. Cheneval,L. Balmelli,P. Prandoni,J.Kovacevic, andM. Vetterli,“In-teractiveDSPeducationusingJava,” in Proc. IEEE Int. Conf. Acoustics,Speech, and Signal Processing, vol. 3, 1998,pp.1905–1908.

[11] A. Spanias,A. Constantinou,J.Foutz,andF. Bizuneh.An onlinesignalprocessinglaboratory.presentedat 1st Signal ProcessingEducationWorkshop.[Online]. Available: http://spib.ece.rice.edu/DSP2000/pro-gram.html

[12] J.Shaffer, J.Hamaker,andJ.Picone,“Visualizationof signalprocessingconcepts,”in Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Pro-cessing, vol. 3, 1998,pp. 1853–1856.

[13] B. L. Evans,L. J. Karam,K. A. West,andJ. H. McClellan,“Learningsignalsandsystemswith mathematica,”IEEE Trans. Educ., vol. 36,pp.72–78,Feb. 1993.

[14] J. GoslingandH. McGilton. (1996)The Java LanguageEnvironment:A WhitePaper.[Online].Available:http://java.sun.com/docs/white/lan-genv/

[15] R. Martti andK. Matti, “An interactive DSP tutorial on the web,” inProc. IEEE Int. Conf. Acoustics, Speech, and Signal Processing, vol. 3,1997,pp. 2253–2256.

[16] C. Ulmer. “How to use PEZ”. [Online]. Available:http://www.ece.ubc.ca/verb1~1edc/466/pezdoc/Demos/use.html

[17] C.VignatandS.Valléry.ZeroPole:A new tool for teachingfilter theoryand design.presentedat 1st Signal ProcessingEducationWorkshop.[Online]. Available:http://spib.ece.rice.edu/DSP2000/

[18] MATLAB Server, Mathworks, Inc.. [Online]. Available:http://www.mathworks.com/products/webserver/

[19] WebCTHomepage[Online]. Available:http://www.webct.com/


[20] A. C. Bovik. EE 371R Digital Image and Video ProcessingCourseNotes.[Online]. Available:http://live.ece.utexas.edu/class/ee371r

[21] , Handbook of Image and Video Processing, 1st ed,ser. Commu-nications,Networking, andMultimediaSeries. New York: Academic, Mar. 2000.

[22] , EE 381-K Digital Signal ProcessingCourseNotes. [Online].Available:http://live.ece.utexas.edu/class/ee381k

[23] LABVIEW homepage[Online]. Available:http://www.ni.com/labview[24] L. K. WellsandJ.Travis,LabVIEW for Everyone, 1sted. UpperSaddle

River, NJ : Prentice-Hall,1997.[25] R. H. Bishop, LabVIEW Student Edition 6i, 1st ed. Upper Saddle

River, NJ: Prentice-Hall,2001.[26] Labview User Manual, 1996.[27] BridgeVIEW and LabVIEW IMAQ Vision for G reference manual, 1996.[28] G. C. Panayi,“Implementationof Digital ImageProcessingFunctions

UsingLabVIEW,” M.S. thesis,Dept.of ElectricalandComputerEngi-neering,Univ. of Texasat Austin , May 1999.

[29] The SIVA Demonstration Gallery [Online]. Available:http://live.ece.utexas.edu/class/siva

[30] MATLAB homepage[Online]. Available:http://www.mathworks.com[31] MathWorks Inc. (1999, May) Getting Started with

MATLAB—Version 5. [Online] http://www.mathworks.com/ac-cess/helpdesk/help/techdoc/learn_matlab/gs_colle.shtml

[32] The Mathworks Inc.. (1996, Dec.) MATLAB Signal Pro-cessing Toolbox. [Online] http://www.mathworks.com/ac-cess/helpdesk/help/toolbox/signal/signal.shtml

[33] The Mathworks Inc.. (1997, June) Building GUIs WithMATLAB-Version 5. [Online] http://www.mathworks.com/ac-cess/helpdesk/help/techdoc/creating_guis/creating_guis.shtml

[34] M. A. Tekalp,Digital Video Processing, 1st ed. UpperSaddleRiver,NJ : Prentice-Hall, 1995.

[35] Wavelab 802 [Online]. Available: http://www-stat.stan-ford.edu/verb1~1wavelab/

Umesh Rajashekar (S’97) received the B.E. degreein electronicsandcom-municationengineeringfrom the KarnatakaRegional EngineeringCollege,Surathkal,India, in July 1998 and the M.S. degree from the University ofTexas at Austin (UT-Austin) in August 2000. He is currently pursuingthePh.D. degree at the Departmentof Electrical and ComputerEngineeringatUT-Austin.

Sincespring2000,he hasbeena ResearchAssistantin the LaboratoryforImageandVideo Engineeringat UT-Austin, wherehe is investigatingimagestatisticsat point of gaze.His interestsalso includedevelopingdidactic toolsfor education.

Mr. Rajashekarwasawardedthe2000–2001TexasTelecommunicationsEn-gineeringConsortiumGraduateFellowshipfrom theUniversityof Texas.

George C. Panayi wasbornin Larnaca,Cyprus,onNovember13,1972.Afterhighschool,hereceivedtheHigherNationalDiplomain electricalengineeringfrom the Higher TechnicalInstitute in Nicosia,Cyprus.He received the B.S.degreein computersciencewith honorsandtheM.S.degreein engineeringfromtheUniversityof TexasatAustin (UT-Austin) in 1997and1999,respectively.

In July 1993,he joined the Cyprusarmy. In the army, he hadfour monthsof officer’s training in Greeceandthenserved asa Second-Lieutenantfor thesupplyandtransportationdivision.SinceJuly1997,hehasbeenaffiliatedwiththeLaboratoryfor Vision Systemsat UT-Austin.

Frank P. Baumgartner is currentlypursuingtheB.S.E.E.degreeattheUniver-sity of Texasin Austin (UT-Austin),whereheis focusingon imageandvideoprocessing.

He is currentlya ResearchAssistantin the Laboratoryfor Vision Systems(LVS) at UT-Austin. There,he developsvideo processingtools for usein alearningenvironment.His interestsincludethehumanpsychovisualsystem,ma-chinevision,anddevelopmentof effective learningtools.

Alan C. Bovik (S’80–M’80–SM’89–F’96) received the B.S. degreein com-puterengineeringandthe M.S. andPh.D.degreesin electricalandcomputerengineeringfrom theUniversityof Illinois, Urbana-Champaign,in 1980,1982,and1984,respectively.

Duringspring1992,heheldavisitingpositionin theDivisionof AppliedSci-ences,HarvardUniversity, Cambridge,MA. He is currentlytheRobertParkerCentennialEndowed Professorin the Departmentof ElectricalandComputerEngineeringand the Director of the Laboratoryfor Imageand Video Engi-neering(LIVE) in the Centerfor PerceptualSystems,theUniversity of Texasat Austin (UT-Austin).He is alsoa frequentconsultantto legal, industrial,andacademicinstitutions.He haspublishedmorethan300technicalarticlesin theresearchareasof digital video,imageprocessing,andcomputationalaspectsofbiologicalvisualperception.Heholdstwo U.S.patents.Heis alsotheeditor/au-thor of the Handbook of Image and Video Processing (New York: Academic,2000).HehasbeenaMemberof theEditorialBoardfor numerousprofessionalsocietypublications,includingPattern Recognition (1988–present),The Journalof Visual Communication and Image Representation (1992–1995),GraphicalModels and Image Processing (1995–1998),Pattern Analysis and Applications(1997–1998),andReal-Time Imaging (2000–present).

Dr. Bovik is a registeredProfessionalEngineerin theStateof Texas.Hehasbeenatwo-timeHonorableMentionwinnerof theinternationalPatternRecog-nition SocietyAward for OutstandingContribution in 1988and1993.He re-ceivedtheInstituteof ElectricalandElectronicsEngineering(IEEE)SignalPro-cessingSocietyMeritoriousServiceAwardin 1998,wasnamedDistinguishedLecturerof the IEEE SignalProcessingSociety, andreceived the IEEE ThirdMillennium Medal andthe University of TexasEngineeringFoundationHal-liburtonAward,all in 2000.In addition,hehasservedontheBoardof Governorsof theIEEE SignalProcessingSocietyfrom 1996to 1998andastheFoundingGeneralChairmanof the First IEEE InternationalConferenceon ImagePro-cessing,Austin,TX, in November1994.HehasservedasAssociateEditorof theIEEETRANSACTIONSON SIGNAL Processing(1989–1993),SteeringCommitteeMemberof theIEEETRANSACTIONSON IMAGE PROCESSING(1991–1995),As-sociateEditor of the IEEE SIGNAL PROCESSINGLETTERS (1993–1995),Ed-itor-in-Chiefof theIEEETRANSACTIONSON IMAGE PROCESSING(1996–2001),andEditorialBoardMemberof thePROCEEDINGSOFTHE IEEE(1998–present).

ieee transactions on education, vol. 45, no. 4, … · matlab and labview are used as the...

Documents