Meridian, a 2016 Netflix release, is available for uncompressed download for engineers, production artists, or anyone else who wants to practice their media processing skills on studio production quality material. Many thanks are due to Netflix for releasing this title under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License
(CC) Netflix licensed under CC BY-NC-ND 4.0
The film is described as “cerebral, full of suspense, chilling, ominous.” I don’t know about the story, but those words certainly describe the process of obtaining and using this film’s downloadable media assets. It is released in multiple color renderings on two different sites, some versions in Interoperable Media Format (IMF) and other encodings wrapped in mp4 packages.
You know with a name like “Interoperable Media Format” that the IMF package is likely to present some barriers to actually access the elementary audio and video media. IMF adds a layer of packaging on top Material eXchange Format (MXF), which by itself is always a challenge to wrap and unwrap.
In this post we’ll enter the Meridian cave armed with only a flashlight and ffmpeg. When we emerge, we’ll have elementary stream files in hand and know what color rendering they belong to.
Subjective video quality evaluation involves people making judgments based on observations and experience. Some professional “golden eyes” do this as part of their job.
Now you can test your own skills in a standards-based video quality assessment campaign that is currently being conducted as a web application. Keep reading to learn more about subjective video quality evaluation, or visit https://pqtest.cascadestream.com to learn by doing. There is an incentive to participate which we talk about in the last paragraph.
What is the quality of this video picture?
Short version: please go to https://pqtest.cascadestream.com and participate in a subjective picture quality evaluation test session. Running now through March 6, 2018.
It’s the evening and the dishes are done. You are relaxing with your favorite person, starting to watch your new favorite drama on your big screen, and Ugh. The picture quality looks terrible! How does this happen? How can we fix it?
Your streaming content service provider actually does care about picture quality. Often the problem is that the process they use to create the compressed video distribution files you see is unattended, and without a human checking every program encoded at every bit rate they simply don’t have the data on picture quality.
You Can Help!
A number of researchers and developers are working on computational models (i.e. automated) that can evaluate video on a human perceptual quality scale. This is a really hard problem but the algorithms are getting better every year.
These researchers and developers need to know the true quality of video in order for them to improve the accuracy of their models. But who can tell them the true quality? Answer: you.
You tell them if it looks good or bad
Cascade Stream is currently hosting a campaign of picture quality evaluation test sessions to determine the true quality of a number of video sequences with various distortions applied to them. We are seeking volunteers to participate and score the video pictures. Your scores will be combined with those of many others resulting in Mean Opinion Scores (MOS) that are the true measurement of quality for those video sequences and distortions. By definition, according to standard recommended practices for subjective video quality evaluation, what you say is what it is.
Do you feel empowered?
You are. Please be an official observer in our picture quality evaluation campaign. To participate, just go to https://pqtest.cascadestream.com. It’s like a focus group that you can do from your home or office. Everything you need to know is on that site. The test session takes about half an hour.
This campaign runs through March 6, 2018 so don’t delay. The library of video pictures, distortions, and quality scores will become a useful resource for developers.
The next time you are watching your favorite new show and it looks great, you may have an automated quality evaluation model that was improved by your scores to thank for it.
Some decades ago I arrived in Portland to relaunch an engineering career after a short, mistaken move to Westchester County, New York. One of my first actions getting set up in Oregon was to build my own desk. There was no money to buy one because I had just returned money for moving expenses to my previous employer.
This desk is still my “daily driver” at Cascade Stream HQ. It is a constant in a professional evolution full of challenges, accomplishments, and setbacks too, like anyone’s career. Another constant is my enjoyment of the practice of engineering development, creating solutions from innovation, or at least concentrated time sitting at that desk working to make a design as simple as it can be to get the job done.
Cascade Stream is now six months old, according to state business registration records. But its real origin dates back to the mid 2000’s as some readers will recall: we talked about this. We launched last year (and not 7 or 10 years ago) because it was a moment when two clients were open to working with a new company. They are
- A leading company in test and measurement
- A semiconductor manufacturer who has redefined low power, high performance computing with a massively parallel processor architecture
These relationships have allowed us to flex our capabilities in the video domain and accomplish some interesting objectives.
The last six months have also included time spent at the College of Engineering and Computer Science at Portland State University learning about software development for web applications. This is another tool in the toolbox to enable cloud-based solutions for video streaming applications.
I’m in school again now, this time at the Portland Community College Small Business Development Center, working on this business.
The next six months will bring new clients and new projects. I will make another trip to the NAB convention in April after countless past visits, but the first for Cascade Stream. The drive for interesting new projects in video keeps it fresh and alive.
Thanks for reading, and thank you for your support.
The International Telecommunications Union publishes Recommendation ITU-R BT.709 which specifies how image data is coded for transmission as video. This includes a specification for the representation of color which applies to almost all modern video transmission including broadcast, cable/satellite, or over the internet (but not including emerging practices for high dynamic range/wide color gamut).
We conventionally understand any color to be a combination of red, green, and blue primaries, but the photoreceptors in the human eye do not function on the basis of these primaries, and BT.709 does not include the first principles needed to understand color representation. It only offers the following specification:
– Red (R)
– Green (G)
– Blue (B)
||Assumed chromaticity for
equal primary signals
In this post, we take a deeper look at Publication No. 15 of the Commission Internationale de l’Eclairage, CIE 15:2004, which includes the 1931 chromaticity specification, still the defacto standard for quantifying color in a camera and display device-invariant way.
Motion estimation in video processing refers to finding a sub-block in a reference picture that most accurately resembles or predicts a block in a target picture. The horizontal and vertical displacement between the target block and the reference block is represented by a two dimensional motion vector.
Spatial domain block matching techniques are most commonly used for motion estimation in video compression encoders and other video processing applications. They work by computing a distance criterion value between a candidate reference block and the target block, according to some candidate motion vector. A common distance criterion function is sum of absolute differences (SAD). The computations need to be repeated for every candidate motion vector, and the one with the smallest distance criterion value is taken as the correct motion vector.
The phase correlation approach uses a frequency domain transformation to find the motion vector in a single iteration. In some applications it could obtain the motion vector with less computations than the spatial domain block matching approach. It may also find the motion vector more accurately when there are extraneous differences between target and reference, such as a different illumination levels or image noise.
We studied motion estimation by phase correlation and wrote a notebook article on it, which you can see on this Jupyter notebook page.
We started with a known test image from which we could see the true motion. Here are the first two frames of the pedestrian_area test clip: