Picture Quality Analysis System

The Tektronix PQA600C Picture Quality Analysis (PQA) System is based on the concepts of the human vision system, and provides a suite of repeatable, objective quality measurements that closely correspond with subjective human visual assessment. These measurements provide valuable information to engineers working to optimize video compression and recovery, and maintaining a level of common carrier and distribution transmission service to clients and viewers.

Compressed Video Requires New Test Methods

The true measure of any television system is viewer satisfaction. While the quality of analog and full-bandwidth digital video can be characterized indirectly by measuring the distortions of static test signals, compressed television systems pose a far more difficult challenge. Picture quality in a compressed system can change dynamically based on a combination of data rate, picture complexity, and the encoding algorithm employed. The static nature of test signals does not provide true characterization of picture quality.

Human viewer testing has been traditionally conducted as described in ITUR Rec. BT.500-11. A test scene with natural content and motion is displayed in a tightly controlled environment, with human viewers expressing their opinion of picture quality to create a Differential Mean Opinion Score, or DMOS. Extensive testing using this method can be refined to yield a consistent subjective rating.

However, this method of evaluating the capabilities of a compressed video system can be inefficient, taking several weeks to months to perform the experiments. This test methodology can be extremely expensive to complete, and often the results are not repeatable. Thus, subjective DMOS testing with human viewers is impractical for the CODEC design phase, and inefficient for ongoing operational quality evaluation. The PQA600C provides a fast, practical, repeatable, and objective measurement alternative to subjective DMOS evaluation of picture quality.

System Evaluation

The PQA600C can be used for installation, verification, and troubleshooting of each block of the video system because it is video technology agnostic: any visible differences between video input and output from processing components in the system chain can be quantified and assessed for video quality degradation. Not only can CODEC technologies be assessed in a system, but any process that has potential for visible differences can also be assessed. For example, digital transmission errors, format conversion (i.e. 1080i to 480p in set-top box conversions), analog transmission degradation, data errors, slow display response times, frame rate reduction (for mobile transmission and videophone teleconferencing), and more can all be evaluated.

How It Works

User Interface of PQA600C

The PQA600C takes two video files as inputs: a reference video sequence and a compressed, impaired, or processed version of the reference. First, the PQA600C performs a spatial and temporal alignment between the two sequences, without the need for a calibration stripe embedded within the video sequence. Then the PQA600C analyzes the quality of the test video, using measurements based on the human vision system and attention models, and then outputs quality measurements that are highly correlated with subjective assessments.

The results include overall quality summary metrics, frame-by-frame measurement metrics, and an impairment map for each frame. The PQA600C also provides traditional picture quality measures such as PSNR (Peak Signal-to-Noise Ratio) as an industry benchmark impairment diagnosis tool for measuring typical video impairments and detecting artifacts.

Each reference video sequence and test clip can have different resolutions and frame rates. This capability supports a variety of repurposing applications such as format conversion, DVD authoring, IP broadcasting, and semiconductor design. The PQA600C can also support measurement clips with long sequence duration, allowing a video clip to be quantified for picture quality through various conversion processes.

Prediction of Human Vision Perception

PQA

PQA600C measurements are developed from the human vision system model and additional algorithms have been added to improve upon the model used in the PQA600C200/300. This new extended technology allows legacy PQR measurements for SD while enabling predictions of subjective quality rating of video for a variety of video formats (HD, SD, CIF, etc.). It takes into consideration different display types used to view the video (for example, interlaced or progressive and CRT or LCD) and different viewing conditions (for example, room lighting and viewing distance).

A model of the human vision system has been developed to predict the response to light stimulus with respect to the following parameters:

This model has been calibrated, over the appropriate combinations of ranges for these parameters, with reference stimulus-response data from vision science research. As a result of this calibration, the model provides a highly accurate prediction.

The graphs below are examples of scientific data regarding human vision characteristics used to calibrate the human vision system model in the PQA600C. Graph (A) shows modulation sensitivity vs. temporal frequency, and graph (B) shows modulation sensitivity vs. spatial frequency. The use of over 1400 calibration points supports high-accuracy measurement results.

Graph A.

Graph B.

Picture (C) is a single frame from the reference sequence of a moving sequence, and picture (D) is the perceptual contrast map calculated by the PQA600C. The perceptual contrast map shows how the viewer perceives the reference sequence. The blurring on the background is caused by temporal masking due to camera panning and the black area around the jogger shows the masking effect due to the high contrast between the background and the jogger. The PQA600C creates the perceptual map for both reference and test sequences, then creates a perceptual difference map for use in making perceptually based, full-reference picture quality measurements.

Image C.

Image D.

Comparison of Predicted DMOS with PSNR

In the examples, Reference (E) is a scene from one of the VClips library files. The image Test (F), has been passed through a compression system which has degraded the resultant image. In this case, the background of the jogger in Test (F) is blurred compared to the Reference image (E).

Image E.

Image F.

A PSNR measurement is made on the PQA600C of the difference between the Reference and Test clip. The highlighted white areas of PSNR Map (G) shows the areas of greatest difference between the original and degraded image.

Another measurement is then made by the PQA600C, this time using the Predicted DMOS algorithm and the resultant Perceptual Difference Map for DMOS (H) image is shown. Whiter regions in this Perceptual Contrast Difference map indicate greater perceptual contrast differences between the reference and test images.

Image G.

Image H.

In creating the Perceptual Contrast Difference map, the PQA600C uses a human vision system model to determine the differences a viewer would perceive when watching the video.

The Predicted DMOS measurement uses the Perceptual Contrast Difference Map (H) to measure picture quality. This DMOS measurement would correctly recognize the viewers perceive the jogger as less degraded than the trees in the background. The PSNR measurement uses the difference map (G) and would incorrectly include differences that viewers do not see.

Attention Model

The PQA600C Opt. BAS and Opt. ADV (not included) also incorporate an Attention Model that predicts focus of attention. This model considers:

Attention Map Example

These attention parameters can be customized to give greater or less importance to each characteristic. This allows each measurement using an attention model to be user-configurable. The model is especially useful to evaluate the video process tuned to the specific application. For example, if the content is sports programming, the viewer is expected to have higher attention in limited regional areas of the scene. Highlighted areas within the attention image map will show the areas of the image drawing the eye's attention.

Artifact Detection

Artifact detection reports a variety of different changes to the edges of the image:

Artifact Detection Settings

They work as weighting parameters for subjective and objective measurements with any combination. The results of these different measurement combinations can help to improve picture quality through the system.

For example, artifact detection can help answer questions such as: "Will the DMOS be improved with more de-blocking filtering?" or, "Should less prefiltering be used?"

If edge-blocking weighted DMOS is much greater than blurring-weighted DMOS, the edge-blocking is the dominant artifact, and perhaps more deblocking filtering should be considered.

In some applications, it may be known that added edges, such as ringing and mosquito noise, are more objectionable than the other artifacts. These weightings can be customized by the user and configured for the application to reflect this viewer preference, thus improving DMOS prediction.

Likewise, PSNR can be measured with these artifact weightings to determine how much of the error contributing to the PSNR measurement comes from each artifact.

The Attention Model and Artifact Detection can also be used in conjunction with any combination of picture quality measurements. This allows, for example, evaluation of how much of a particular noticeable artifact will be seen where a viewer is most likely to look.

Comprehensive Picture Quality Analysis

The PQA600C provides Full Reference (FR) picture quality measurements that compare the luminance signal of reference and test videos. It also offers some No Reference (NR) measurements on the luminance signal of the test video only. Reduced Reference (RR) measurements can be made manually from differences in No Reference measurements. The suite of measurements includes:

Configure Measure Dialog

The PQA600C supports these measurements through preset and user-defined combinations of display type, viewing conditions, human vision response (demographic), focus of attention, and artifact detection, in addition to the default ITU BT-500 conditions. The ability to configure measurement conditions helps CODEC designers evaluate design trade-offs as they optimize for different applications, and helps any user investigate how different viewing conditions affect picture quality measurement results. A user-defined measurement is created by modifying a preconfigured measurement or creating a new one, then saving and recalling the userdefined measurement from the Configure Measure dialog menu.

Edit Measure Dialog

Easy-to-Use Interface

The PQA600C has two modes: Measurement and Review. The Measurement mode is used to execute the measurement selected in the Configure Dialog. During measurement execution, the summary data and map results are displayed on-screen and saved to the system hard disk. The Review mode is used to view previously saved summary results and maps created either with the measurement mode or XML script execution. The user can choose multiple results in this mode and compare each result side by side using the synchronous display in Tile mode. Comparing multiple results maps made with the different CODEC parameters and/or different measurement configurations enables easy investigation of the root cause of any difference.

Multiple Result Display

Resultant maps can be displayed synchronously with the reference and test video in a summary, six-tiled, or overlaid display. In Summary display, the user can see the multiple measurement graphs with a barchart along with the reference video, test video, and difference map during video playback. The user also can select two measurement results on a graph with auto time shifting that absorbs the timing difference at the content capture tom compare two measurement results intuitively. Summary measures of standard parameters and perceptual summation metrics for each frame and overall video sequence are provided.

In Six-tiled display, the user can display the 2 measurement results side by side. Each consists of a reference video, test video, and difference map to compare to each other.

In Overlay display, the user can control the mixing ratio with the fader bar, enabling co-location of difference map, reference, and impairments in test videos. Error logging and alarms are available to help users efficiently track down the cause of video quality problems. All results, data, and graphs can be recalled to the display for examination.

Graph Display

Six-tiled Display

Overlay Display

Automatic Temporal/Spatial Alignment

The PQA600C supports automatic temporal and spatial alignment, as well as manual alignment.

The automatic spatial alignment function can measure the cropping, scale, and shift in each dimension, even across different resolutions and aspect ratios. If extra blanking is present within the standard active region, it is measured as cropping when the automatic spatial alignment measurement is enabled.

Auto Spatial Alignment

The spatial alignment function can be used when the reference video and test video both have progressive content. In the case where the reference video and test video has content with different scanning (interlace versus progressive or vice versa), the full reference measurement may not be valid. In the case where the reference video and test video both have interlaced content, the measurement is valid when spatial alignment is not needed to be set differently from the default scale and shift.

Region of Interest (ROI)

There are two types of spatial/temporal Region of Interest (ROI): Input and Output. Input ROIs are used to eliminate spatial or temporal regions from the measurement which are not of interest to the user. For example, Input Spatial ROI is used when running measurements for reference and test videos which have different aspect ratios. Input Temporal ROI, also known as temporal sync, is used to execute measurements just for selected frames and minimize the measurement execution time.

Output Spatial ROI

Output ROIs can be used to review precalculated measurement results for only a subregion or temporal duration. Output Spatial ROI is instantly selected by mouse operation and gives a score for just the selected spatial area. Its an effective way to investigate a specific spatial region in the difference map for certain impairments. Output Temporal ROI is set by marker operation on the graph and allows users to get a result for just a particular scene when the video stream has multiple scenes. It also allows users to provide a result without any influence from initial transients in the human vision model. Each parameter can be embedded in a measurement for the recursive operation.

Automation Test with XML Scripting

In the CODEC debugging/optimizing process, the designer may want to repeat several measurement routines as CODEC parameters are revised. Automated regression testing using XML scripting can ease the restrictions of manual operation by allowing the user to write a series of measurement sequences within an XML script. The script file can be exported from or imported to the measurement configuration menu to create and manage the script files easily. Measurement results of the script operation can be viewed by using either the PQA600C user interface or any spreadsheet application that can read the created .csv file format as a summary. Multiple scripts can be executed simultaneously for faster measurement results.

Script Sample

Import/Export Script

Result File Sample

HDMI Compliant with HDCP Support

SD/HD/3G SDI, HDMI Compliant Interface and IP Interface

An SD/HD SDI interface and IP interface enable both generation and capture of SDI video and IP video. The HDMI compliant with HDCP support allows the user to directly capture the HDCP encrypted contents from the consumer instruments such as Blu-ray player and Set Top Box without hassle. This is beneficial for comparing the performance in multiple units/ models or monitoring the picture quality of end to end broadcast chain including the STB output at home.

There are three modes of simultaneous generation capture operation that can be performed on all video formats except 1080p 50/59/60 formats: generation and capture, 2-channel capture, and 2-channel generation.

Simultaneous Generation/Capture

Simultaneous Generation and Capture

Simultaneous generation and capture lets the user playout the reference video clips directly from the PQA600C into the device under test. The test output from the device can then be simultaneously captured by the PQA600C. The real time up / down converter could be inserted in the video signal path at generation or capture operation to evaluate an instrument with up / down conversion process.

Simultaneous 2-Channel Capture

Simultaneous 2-Channel Capture

Simultaneous 2-channel capture lets the user capture two live signals to use as reference and test videos in evaluating the device under test in operation. To accommodate equipment processing delay that may be present in the system, the user can use the Delay Start function when capturing video. Using Delayed Start minimizes the number of unused overhead frames in the test file and enables faster execution of the auto temporal alignment in the measurement.

Simultaneous 2-Channel Generation

Simultaneous 2-Channel Generation

Simultaneous 2-channel generation capability, available only in SDI/HDMI interface selection, supports three types of subjective testing with one display. Swap-channel capability will exchange reference and test video sources in a frame to help the user to figure out the difference without moving the focus point of their eyes.

Side-by-side display arranges the video output from the regions in the reference and test video lining up in a row. The Wipe display takes the left region of reference video and the right region of the test video and merges them into a single video output seamlessly.

Side-By-Side Display

Wipe Display

IGMP Support

IGMP User Interface

In any modes, the user can select the Cross Interface configuration such as generating from SDI/HDMI and capturing from IP or vice versa. IGMP support in IP capture will make stream selection simple at multicast streaming. The compressed video file captured through IP will be converted to an uncompressed file by an internal embedded decoder.