Video Quality Analyzer

In the main menu, the most frequently selected choice displays the list of processed videos. In this list, each video is represented by a unique number, its test campaign (that enables to gather several results), its file name, its full path, the description of its video and audio streams (codec, dimensions, frame rate, duration, etc...), its video quality score, statistics on video quality (min, max, standard deviation) and parameters used for synchronization. For each video, several links enable you to display the quality curves, the bitrate curve, the characteristic frames. Some other links permit to draw various representations of the perceived quality versus the bitrate. At last, some links allow to generate perceived video quality analysis reports in various formats (CSV, TXT, HTML).

In the web interface, all curves are interactive: if you click on one point of the curve, it will put this point at the center and zoom by a factor of 2. And if you drag your mouse to select a portion of the curve, it will zoom on this selection. If the frame corresponding to the center of the curve is a characteristic frame, then this frame and its reference frame will be displayed below the curve. And if the distorted frame contains areas that have been distorted beyond an acceptability threshold defined by the user, then these areas will be boxed, so that you can find locate easily. The color of the box indicates the level of distortion (the clearer the color, the higher the distortions). And if the boxes are annoying, you can make them invisible using a single click.

If several distorted videos have the same reference video, then you can draw all their curves on the same figure, in order to compare the effects of different encoders or different bitrates on a given video content. You can also get graphical representations of the perceived quality versus the bitrate. This will enable you to determine the bitrate which is required to get a given quality level.

These different features help the user during results analysis. Indeed, generally we consider that there are 5 levels of analysis. Indeed, let's suppose that you measured the perceived video quality of a video file.

• Level 1: First you'll have a look at the mean quality score. VQA shows it. If you consider that this score is good (or good enough), then you're lucky and your work is finished. But most of the time, you'll think that this score is not good or not good enough (and that you can certainly improve it).
• Level 2: So you'll want to have a look at the quality curve, to see if you always have a constant quality (all along the video) or if you have peaks of quality loss at certain moments. You just have to click on a link and VQA will display this curve.
• Level 3: You detect some peaks and you want to see the corresponding frames, to evaluate the quality of these frames by yourself. Just click on a peak and VQA will display the distorted and reference frames corresponding to this peak. It will also display the curve centered around this peak.
• Level 4: You will then explore the frames that correspond to peaks and try to find the most distorted zones. Thanks to the boxes, VQA shows these zones to you.
• Level 5: Generally, when you have a look at the most distorted zones, you make conclusions like "OK that's very distorted because it was a fine texture for which small details were removed by compression". But the difficulty is that, doing this way, you will only have qualitative conclusions and no quantitative data about the reason of these quality losses. VQA does better, thanks to content classification!

In parallel of perceived video quality measurement and bitrate measurement, VQA also divides the image into several zones. Each zone is classified into several categories like "Homogenous area", "Fine texture", "Horizontal contour", "Vertical contour", "Diagonal contour", "Strong texture". Each category is even subdivided into two types: "low luminance" and "high luminance".

For each category, VQA will measure the number of pixels in this category and the perceived video quality of the corresponding zones.

This enables to display the statistics required for level 5 (see above). Indeed, VQA indicates the proportion of each category (the number pixels from the video belonging to each category), the quality of each category and which category suffers from the most important distortions (in the video and, by normalization, in a video which would contain the same number of pixels in each category). If you work on encoders R&D or if you want to finely tune an encoder, this will enable you to adapt the quantization coefficients to maximize perceived video quality!

This technology is patented and exclusively available in AccepTV's Video Quality Analyzer.

VQA can also measure audio quality. However, the implemented audio quality metric is simpler than the video quality metric. The audio quality metric is similar to an "improved PSNR" (improvement due to audio gain compensation).

When audio processing is enabled, VQA performs audio synchronization, audio quality measurement, instant audio bitrate measurement and mean audio bitrate computation.

VQA also enables to automatically encode audio video files with the most common containers (TS, MP4, AVI, etc.) and the most popular audio video codecs (H.264, MPEG-2, AAC, MP3, etc.).

And thanks to its advanced perceived video quality measurement technologies, VQA can even automatically determine the optimal video bitrate to reach the quality level you want!

Of course, if you prefer using predefined video bitrates without adjusting them, you can still use VQA's encoding's watch folder to automatically encode/transcode your audio video files.

Encoding is based on a watch folder. When a new audio video file is copied or moved to the watched folder or any of its subfolders, it is encoded using parameters (codecs, container, etc.) which are defined by the user and specified for each folder. Therefore, to encode a given file with given parameters, you just have to copy or move this file to the appropriate folder. And when encoding is finished, a user-defined script can be run (to move the output file to a given folder or to delete the input file, for example).

VQA contains many other features, among which:

• Tasks Manager: when you click on a button, you don't directly perform an action but in fact you add a task in a queue. This enables you to continue loading files, synchronizing them and measuring them, even if VQA is already working.
• Watch Folder: it enables to detect every file appearing (copied) in a folder (or its subfolders). For each new file, several rules (based on filenames, especially on end of filenames) can be applied to determine the possible names of a reference file. If a reference file is found, the quality measurement is launched. The Watch Folder enables to monitor a file-based encoding process using VQA (for VOD, for example)
• Automation methods: to facilitate the use of VQA with an important number of files, VQA can load scripts or be used in command line (with arguments).
• Magnifying glass and signal profiles: to explore the images at pixel level, view the Y, U and V values of each pixel and to compare two frames coming from two different files, a magnifying glass is available, along with horizontal and vertical signal profiles displaying. These windows show the image content in the neighborhood of the mouse cursor's position.
• Video Cutter: a tool to cut both the reference video and the measured video into several video clips, using start/stop markers for the reference video and using synchronization for the measured video. Video Cutter simplifies the use of a long duration video which contains different contents (for example: sport, movie, news, cartoon, concert, etc.) by splitting this long duration video in several video clips. Each video clip will contain a single content type (one video clip for sport, one video clip for movie, etc). Thanks to it, you'll be able to have one quality score for each content type (one quality score for sport, one quality score for movie, etc).

VQA includes AccepTV's full-reference video quality metrics but also Netflix's VMAF metric.

By using VMAF in VQA, you can compute the mean VMAF value of each processed video but you can also:

• Compute VMAF for any supported file format (*.mp4, *.avi, *.ts, etc.) and not only for raw YUV files
• Compute VMAF for videos which are not temporally synchronized thanks to VQA's automatic synchronization
• Compute VMAF on video files of any duration (no limitation)
• Get the VMAF value of each processed video frame
• Automatically see which video frames have the minimum and maximum VMAF values in a video file
• Get curves of VMAF values with respect to time
• Trigger alerts when VMAF scores are too low
• Measure audio quality in parallel, using the "improved PSNR" audio metric

With all these features (and more, like scripting), VQA is a perfect tool to use VMAF!

VQA includes the following metrics:

• Our full-reference perceptual video quality metrics (3 metrics, dedicated to MPEG-2, H.264 and HEVC)
• MSE & PSNR
• Lipsync metric
• VMAF

VQA also includes these other metrics:

• PSNR-HVS: Peak Signal-to-Noise Ratio taking into account Contrast Sensitivity Function (CSF)
• PSNR-HVS-M: Peak Signal-to-Noise Ratio taking into account Contrast Sensitivity Function (CSF) and between-coefficient contrast masking of DCT basis functions
• VIFp: Visual Information Fidelity (pixel domain version)
• SSIM: Structural Similarity
• MS-SSIM: Multi-Scale Structural Similarity

For all quality metrics, the reference video and measured video can even have different resolutions and/or different frame rates. For example, VQA can measure the perceived video quality of a distorted video in 720p format at 15 frames per second using a reference video in 1080p format at 25 frames per second.

With all these metrics (and its unique features), VQA is certainly the most complete tool for full-reference video quality measurement.