Tektronix PQA600 DATASHEET

Picture Quality Analysis System

PQA600 Data Sheet

Applications

Features & Beneﬁts

Fast,Accurate,Repeatable,and ObjectivePicture Quality Measurement Predicts DMOS(Differential Mean Opinion Score) based on Human

VisionSystem Model Picture QualityMeasurements can bemade on aVarietyof HD Video

Formats ( 1080i, 720p) andSD VideoFormats (525ior 625i) Makes PictureQuality Comparison acrossDifferent Resolutions fro m

HD toSD, or SD/HDto CIF User-conﬁgurable ViewingCondition andDisplay Modelsfor Reference

and Comparison Attention/Artifact WeightedMeasurement RegionOf Interest (ROI) on MeasurementExecution andReview Automatic Temporal andSpatial Alignment Easy Regression Testing and Automation using XML Scriptingwith

"Export/Import" Filefrom GUI Multiple Results View Options Optional SD/HDSDI Interfacewith Simultaneous Generation/Capture,

2-channel Capture and 2-channel Gene ration with Swap-channel Capability

Preinstalled Sample Reference andTestSequences

CODEC Design Conformance Testing,Transmission Equipme nt, and System Evaluation Digital VideoMastering Video CompressionServices DigitalConsumer Product Developmentand Manufacturing

, Optimization,and Veriﬁcation

Picture Quality Analysis System

The PQA600is the latest-generation Picture QualityAnalyzer built onthe EmmyAward winningTektronixPQA200/300. Basedonthe conceptsof the human vision system, thePQA600 providesa suiteof repeatable,objective quality mea visual assessment. These measurements provide valuable info rmation to engineersworking to optimize videocompression and recovery,and maintaining a level ofcommon carrierand distributiontransmission service to clientsand viewers.

Compressed Video Requires New Test Methods

The truemeasure of anytelevision system isviewer satisfaction. While the qualityof analog andfull-bandwidth digital videocan becharacterized indirectly bymeasuring the distortionsof static testsignals, compressed television systemspose afar more difﬁcultchallenge. Picturequality in a compressed systemcan change dynamicallybased on acombination of data rate staticnature of te st signals doesnot providetrue characterizationof picture quality.

Human viewertesting has been traditionallyconducted as described in ITU-R Rec. BT.500-11. Atest scene with natural content and motion is displayed ina tightly controlled environment, with human viewers expressing theiropinion of picture quality to create a DifferentialMean Opinion Score, orDMOS. Extensive testing using this method can be reﬁned to evaluatingthe capabilities of a compressed video system canbe inefﬁcient, taking severalwee ks to months to perform the experiments. This test methodology can be extremelyexpensive tocomplete, andoften theresults are notrepeatable. Thus, subjectiveDMOS testing withhuman viewers is impracticalfor the CODEC designphase, and inefﬁcient forongoing operati repeatable, andobjective measurement alternative to subjective DMOS evaluation ofpicture quality.

surements that closelycorrespond with subjective human

, picturecomplexity,and the encodingalgorithm employed. The

yield aconsistent subjective rating. However, this method of

onal qualityevaluation. The PQA600 provides afast, practical,

Data Sheet

UserInterface of PQA600. Showing reference, testsequences, with differencemap and statisticalgraph.

System Eva

luation

The PQA600 can beused for installation, veriﬁcation, andtroublesh ooting ofeach blo

ckof the videosystem becauseit isvideo technology agnostic: any visibledifferences between videoinput and output fromprocessing componentsin the systemchain canbe quantiﬁedand assessedfor video quality degradation. Not only canCODEC technologies beassessed in a system, butany process that has potentialfor visible differencescan also be assessed. Forexample, digital transmissionerrors, format co nversion

0i to480p in set-top box conversions), 3-2pull-down, analog

(i.e. 108 transmission degradation, da ta errors, slowdisplay responsetimes, frame rate reduction(for mobile transmissionand videophone teleconferencing), andmore can allbe evaluated,separately orin anycombination.

How It Works

The P QA600 takes twovideo ﬁles asinputs: areference videosequence and a compressed, impaired,or processed versionof thereference. First, the PQA600performs a spatialand temporal alignmentbetween the two sequences, withoutthe need fora calibration stripeembedded within the

quence. Then the PQA600 analyzes the qualityof the testvideo,

videose using measurementsbased on the humanvision system and attention models, an d then outputsquality measurements thatare highlycorrelated with subjectiveassessments. Theresults include overallquality summary metrics,frame-by-frame measurement metrics,and animpairment map for eachframe. ThePQA600 alsoprovides traditionalpicture qualitymeasures such asP

SNR (PeakSignal-to-Noise Ratio) as anindustry benchmark impairment diagnosistool for measuring typicalvideo impairments and detecting artifacts.

Eachreference video sequenceand testclip canhave different resolution s and framerates. The PQA600 can provide picturequality measurement betweenHD vs SD, SD vsCIF,or anycombination. Thiscapability sup ports

Picture Quality Analysis System

a varietyof repurposing applications such as format conversion, DVD authoring, IPbroadcasting, and semiconductordesign. ThePQA600 can also supportmeasurement clips withlong sequence duration,allowing a video clipto be quantiﬁed for picture quality throughvarious conversion processes.

Prediction of Human Vision Perception

PQA600 measurementsare developed from thehuman vision system model andadditional algorithms have beenadded to improve uponthe model usedin the PQA200/300. Thisnew extended technology allows legacy PQR measurements forSD whileenabling predictions ofsubjective quality ratingof video for a varietyof video formats(HD, SD, CIF, etc.). It takes into consideration different display typesused to view the video(for example, interlacedor progressiveand CRTor LCD) anddifferent viewing conditions (for example, roomlighting and viewin g distance).

A modelof the human visionsystem has beendeveloped to predict the response tolight stimuluswith respectto thefollowing parameters:

Contrast includingSupra-threshold Mean Luminance Spatial Frequency Temporal Frequency Angular Extent Temporal Extent Surround Eccentricity Orientation Adaptation Effects

2 www.tektronix.com

A: Modulation Sensitivity vs. TemporalFrequency

Picture Quality Analysis System— PQA600

C: Reference Picture

B: Modulation Sensitivity vs. SpatialFrequency

This modelhas been calibrated, over theappropriate combinations of

s for these parameters, with reference stimulus-response data from

range visionscience research. As aresult ofthis calibration,the modelprovides a highly accurateprediction.

ptual Contrast Map

D:Perce

The graphsabove areexam ples of scientiﬁcdata regarding humanvision characteristics usedto calibrate the humanvision system model in the PQA600. Graph (A) showsmodulation sensitivity vs. temporal frequency, and graph (B ) shows modulation sensitivity vs. spatial frequency. The

over 1400calibration pointssupports high-accuracy measurement

use of results.

re (C)is a single frame fro m the reference sequence ofa moving

Pictu sequence, andpicture (D)is theperceptual contrast mapcalculated bythe PQA600. The perceptualcontrast mapshows howthe viewerperceives the reference s equence. The blurringon thebackground iscaused by temporal masking due to camera panning and theblack area around the jogger showsthe masking effectdue to the highcontrast betweenthe background

ejogger. The PQA600creates the perceptualmap forboth reference

andth and testsequences, then createsa perceptual differencemap for usein making percep tually based, full-referencepicture quality measurements.

www.tektronix.com 3

Data Sheet

E: Reference

F:Test

Comparison of Predicted DMOS with PSNR

In theexample above, Reference (E)is a scenefrom oneof the VClips libraryﬁles. Theimage Test (F),has been passedthrough a compression system which has degraded the resultant image. In this case the background ofthe joggerin Test (F) is blurred comparedto the Reference image (E). A PSNR measurement ismade onthe PQA600of thedifference

en the Reference and Testclip and the highlighted whiteareas

betwe of PSNR Map (G) shows the areasof greatest difference betweenthe original and deg raded image. Anothermeasurement is then made bythe PQA600, thistime usingthe Predicted DMOSalgorithm and theresultant Perceptual D ifference Map forDMOS (H) imageis shown. Whiter regions

G: PSNR Map

ptual Difference Map forDMOS

H:Perce

in thisPerceptual Contrast Difference mapindicate greater perceptual contrast differencesbetween the referenceand test images. In creating the Perceptual Contrast Differencemap, thePQA600 usesa human vision system model to determinethe differencesa viewer wouldperceive when

ing the video.

watch The Predicted DMOS measurement uses the Perceptual Contrast

rence Map(H) to measure picture quality. ThisDMOS measurement

Diffe wouldcorrectly recognize theviewers perceivethe joggeras lessdegraded than the treesin the background. The PSNR measurement uses the differencemap (G) andwould incorrectly includedifferences that viewers do not see.

4 www.tektronix.com

AttentionMap Example: The joggeris highlighted

Attention Model

ThePQA600 alsoincorporates an AttentionModel thatpredicts focusof attention. Thismodel considers:

Motion of Objects Skin Coloration(to identify people) Location Contrast Shape Size ViewerDistraction due toNoticeable Quality Artifacts

These attention parameters canbe customized to give greateror less importanceto each characteristic. Thisallows each measurementusing an

on modelto beuser-conﬁgurable. Themodel is especiallyuseful to

attenti evaluatethe video process tuned tothe speciﬁcapplication. For example, if the content issports programming, theviewer is expected to have higher attentionin limited reg ional areas of the scene. Highlightedareas within th e attention imagemap will showthe areas of th e image drawing theeye’s attention.

Artifact Detection

Artifact Detection reports avariety of differentchanges tothe edges of the image:

Loss ofEdges or Blurring Addition ofEdges orRinging/Mosquito Noise Rotationof Edges to Verticaland Horizontalor EdgeBlockiness Loss of Edges withinan Image Block or DCBlockiness

They work as weighting parameters for subjective and objective measurements with any combination. The resultsof these different

rement combinationscan help toimprove picture qualitythrough

measu the system.

Picture Quality Analysis System— PQA600

ArtifactDetection Settings

For example,artifact detection canhelp answer questionssuch as: “Will theDMOSbeimprovedwithmorede-blockingﬁltering?” or, “Should less preﬁltering be used ?”

If edge-blockingweighted DMOS ismuch greater thanblurring-weighted DMOS, theedge-b loc king isthe dominant artifact, and perhaps more de-blocking ﬁlteringshould be considered.

In someapplications, itma y be knownthat added edges,such as ringing and m osquito noise, aremore objectionablethan the otherartifacts. These weightingscan becustomized bythe userand conﬁgured forthe application toreﬂectthis viewerpreference, thus improvingDMOS prediction.

Likewise, PSNRcan be measured with these artifact w determine howmuch of theerror contributing tothe PSNR m easurement comes from each artifact.

The Att ention Model andArtifact Detectioncan also be used inconjunction with anycombination of picture quality measurements. This allows,for

le, e valuation of howmuch of aparticular noticeable artifactwill be

examp seen wherea viewer ismost likely tolook.

eightings to

www.tektronix.com 5

+ 11 hidden pages