Several other things I should be doing on a Saturday morning, but I find myself anxious to continue this discussion. Maybe it’s because although multiple industry Best Practice documents talk about correcting Aspect Ratio, none of them discuss the proper way to do it. It could also be my new coffee maker, which I'm hypothesizing has increased my caffeine intake substantially, although I have not increased my coffee intake. Who knows. Anyway, let’s start by recapping Part 1.
Video existed before the pixel. In fact, the NTSC standard itself existed before the pixel, the CCD sensor, digital camera, Internet, personal computer, and the World Wide Web. ITU Rec. 601 standardized uncompressed NTSC and PAL signal sampling, and proper application of that standard produces a digital file containing non-square pixels; in other words, the sampled data stored needs to be corrected in order to most accurately represent the scene recorded IF it is intended to be displayed in a square pixel environment. DV, CIF, SIF, DVD-Video and others like the original h.261 and h.263 standards were all based off from this concept of storing the originally sampled video signal with non-square pixels.
Analog CCTV Cameras & DCCTV Recorders
Digital cameras, such as IP CCTV cameras, are based on the capture, processing, transmission, storage and display of square pixels. Most analog CCTV cameras, and Digital CCTV (DCCTV) recorders that support analog cameras, are designed and configured to capture, transmit, and store non-square pixels in a pixel matrix that is intended to be displayed at a 4:3 (1.333:1) Aspect Ratio. There are a few exceptions though, like Sony’s 960H cameras for instance; however, leveraging those formats requires both a camera and a recorder capable of supporting the format.
DAR, PAR, SAR
If we know that the originally sampled analog source is intended to be displayed at the 4:3 Aspect Ratio, and we know the pixel matrix for the originally sampled and stored non-square pixel data, we can easily compute the proper pixel matrix for display in a square pixel environment. A few definitions:
- Aspect Ratio: The width to height ratio of an image. (ref: SWGDE/SWGIT Digital & Multimedia Evidence Glossary, v2.8, May 27, 2015 - PDF)
- Storage Aspect Ratio (SAR): The width to height ratio of an image as it is stored/sampled.*
- Display Aspect Ratio (DAR): The width to height ratio of an image as it is intended to be displayed.*
- Pixel Aspect Ratio (PAR): The width to height ratio of each individual pixel.*
* - These definitions are the way I personally prefer to define these terms, and may or may not be defined as such elsewhere. They are not currently defined in any forensic DME Best Practices, Glossaries or Guidelines that I am aware of.
If the source uses square pixels, then the SAR and DAR will be the same, and the PAR will be 1:1. Unfortunately, it’s not as simple as that, as some formats and programs do not support Pixel Aspect Ratio (PAR), so they have no way of knowing how the originally sampled data should be displayed. To them, there is no such thing as PAR. The AVI file format is a prime example; regardless of what software you use, it can only guess at how the data is intended to be displayed accurately. It gets worse.
Even if the format and software support PAR, it is often not addressed properly in encoding parameters or is completely overlooked, so the resulting file encoded does not have the necessary PAR information to tell ANY software how to display it accurately.
“You’re really starting to pi** me off Larry.” I totally get it, trust me. I pick nits, it’s what I do. One more thing though…
Interpolation Methodology
There are many interpolation methodologies and algorithms, most of which are designed of course to produce a more visually pleasing picture. In a forensic environment one of our primary goals is to maintain the integrity of the original data. We want to correct the original data by making the non-square pixels square, not by creating new pixels and data that did not exist in the originally sampled scene. Down sampling corrects the original data, while up sampling creates new data. The latter is often used in video production and multimedia software because, depending on the algorithm being used and the original source, it may produce a more visually pleasing image (e.g. softer, smoother, etc.).
“Isn’t that what we're after Larry, a more visually pleasing image?” The answer depends on the questions being asked of the evidence, and whether or not the more visually pleasing method will affect accurate interpretation of the original source. I will readily admit that most often, up sampling solely for the purpose of correcting aspect ratio does not affect accurate interpretation of the original source; but it could, as it does create new data that did not exist in the source.
Furthermore, down sampling typically improves the Signal-to-Noise Ratio (SNR), but I digress.
In the next post we'll delve into the standards a bit more, go over line & sample doubling, and I’ll show a few examples from CCTV recordings. Feel free to share your feedback via the public comments section below or via our private member forums. I look forward to hearing from you. In the meantime, I'm going to go have a cigar and update my to-do list. Be safe out there my friends.