Generation loss

Generation loss is the loss of quality between subsequent copies or transcodes of data. Anything that reduces the quality of the representation when copying, and would cause further reduction in quality on making a copy of the copy, can be considered a form of generation loss. File size increases are a common result of generation loss, as the introduction of artifacts may actually increase the entropy of the data through each generation.

Digital generation loss induced by rotating a JPEG image 90 degrees (from top to bottom) 0, 100, 200, 500, 900, and 2000 times (without using lossless tools)

Analog generation loss

In analog systems (including systems that use digital recording but make the copy over an analog connection), generation loss is mostly due to noise and bandwidth issues in cables, amplifiers, mixers, recording equipment and anything else between the source and the destination. Poorly adjusted distribution amplifiers and mismatched impedances can make these problems even worse. Repeated conversion between analog and digital can also cause loss.

Generation loss was a major consideration in complex analog audio and video editing, where multi-layered edits were often created by making intermediate mixes which were then "bounced down" back onto tape. Careful planning was required to minimize generation loss, and the resulting noise and poor frequency response.

One way of minimizing the number of generations needed was to use an audio mixing or video editing suite capable of mixing a large number of channels at once; in the extreme case, for example with a 48-track recording studio, an entire complex mixdown could be done in a single generation, although this was prohibitively expensive for all but the best-funded projects.

The introduction of professional analog noise reduction systems such as Dolby A helped reduce the amount of audible generation loss, but were eventually superseded by digital systems which vastly reduced generation loss.[1]

According to ATIS, "Generation loss is limited to analog recording because digital recording and reproduction may be performed in a manner that is essentially free from generation loss."[1]

Digital generation loss

Used correctly, digital technology can eliminate generation loss. This implies the exclusive use of lossless compression codecs or uncompressed data from recording or creation until the final lossy encode for distribution through internet streaming or optical discs. Copying a digital file gives an exact copy if the equipment is operating properly. This trait of digital technology has given rise to awareness of the risk of unauthorized copying. Before digital technology was widespread, a record label, for example, could be confident knowing that unauthorized copies of their music tracks were never as good as the originals.

Generation loss can still occur when using lossy video or audio compression codecs as these introduce artifacts into the source material with each encode or reencode. Lossy compression codecs such as Apple ProRes, Advanced Video Coding and mp3 are very widely used as they allow for dramatic reductions on file size while being indistinguishable from the uncompressed or losslessly compressed original for viewing purposes. The only way to avoid generation loss is by using uncompressed or losslessly compressed files; which may be expensive from a storage standpoint as they require larger amounts of storage space in flash memory or hard drives per second of runtime. Uncompressed video requires a high data rate; for example, a 1080p video at 30 frames per second can require up to 370 megabytes per second.[2] Lossy codecs make Blu-rays and streaming video over the internet feasible since neither can deliver the amounts of data needed for uncompressed or losslessly compressed video at acceptable frame rates and resolutions. Images can suffer from generation loss in the same way video and audio can.

Processing a lossily compressed file rather than an original usually results in more loss of quality than generating the same output from an uncompressed original. For example, a low-resolution digital image for a web page is better if generated from an uncompressed raw image than from an already-compressed JPEG file of higher quality.

Techniques that cause generation loss in digital systems

In digital systems, several techniques, used because of other advantages, may introduce generation loss and must be used with caution. However, copying a digital file itself incurs no generation loss—the copied file is identical to the original, provided a perfect copying channel is used.

Some digital transforms are reversible, while some are not. Lossless compression is, by definition, fully reversible, while lossy compression throws away some data which cannot be restored. Similarly, many DSP processes are not reversible.

Thus careful planning of an audio or video signal chain from beginning to end and rearranging to minimize multiple conversions is important to avoid generation loss when using lossy compression codecs. Often, arbitrary choices of numbers of pixels and sampling rates for source, destination, and intermediates can seriously degrade digital signals in spite of the potential of digital technology for eliminating generation loss completely.

Similarly, when using lossy compression, it will ideally only be done once, at the end of the workflow involving the file, after all required changes have been made.

Transcoding

Converting between lossy formats – be it decoding and re-encoding to the same format, between different formats, or between different bitrates or parameters of the same format – causes generation loss.

Repeated applications of lossy compression and decompression can cause generation loss, particularly if the parameters used are not consistent across generations. Ideally an algorithm will be both idempotent, meaning that if the signal is decoded and then re-encoded with identical settings, there is no loss, and scalable, meaning that if it is re-encoded with lower quality settings, the result will be the same as if it had been encoded from the original signal – see Scalable Video Coding. More generally, transcoding between different parameters of a particular encoding will ideally yield the greatest common shared quality – for instance, converting from an image with 4 bits of red and 8 bits of green to one with 8 bits of red and 4 bits of green would ideally yield simply an image with 4 bits of red color depth and 4 bits of green color depth without further degradation.

Some lossy compression algorithms are much worse than others in this regard, being neither idempotent nor scalable, and introducing further degradation if parameters are changed.

For example, with JPEG, changing the quality setting will cause different quantization constants to be used, causing additional loss. Further, as JPEG is divided into 16×16 blocks (or 16×8, or 8×8, depending on chroma subsampling), cropping that does not fall on an 8×8 boundary shifts the encoding blocks, causing substantial degradation – similar problems happen on rotation. This can be avoided by the use of jpegtran or similar tools for cropping. Similar degradation occurs if video keyframes do not line up from generation to generation.

Editing

Digital resampling such as image scaling, and other DSP techniques can also introduce artifacts or degrade signal-to-noise ratio (S/N ratio) each time they are used, even if the underlying storage is lossless.

Resampling causes aliasing, both blurring low-frequency components and adding high-frequency noise, causing jaggies, while rounding off computations to fit in finite precision introduces quantization, causing banding; if fixed by dither, this instead becomes noise. In both cases, these at best degrade the signal's S/N ratio, and may cause artifacts. Quantization can be reduced by using high precision while editing (notably floating point numbers), only reducing back to fixed precision at the end.

Often, particular implementations fall short of theoretical ideals.

Photocopying, photography, video, and miscellaneous postings

Successive generations of photocopies result in image distortion and degradation.[3] It has been documented that successive repostings on Instagram results in noticeable changes.[4] Likewise, repeated postings on YouTube degraded the work.[5][6] Similar effects have been documented in copying of VHS tapes.[5] This is because both services use lossy codecs on all data that is uploaded to them, even if the data being uploaded is a duplicate of data already hosted on the service, while VHS is an analog medium, where effects such as noise from interference can have a much more noticeable impact on recordings.

Generation loss has been the subject of artistic works that exposed the phenomenon.[7] The Generation Loss project was inspired, at least in part, by an earlier effort—"I Am Sitting in a Room" (1969) by Alvin Lucier.[7][8]

See also

Notes
  1. "generation loss". ATIS Telecom Glossary. Alliance for Telecommunications Industry Solutions. September 28, 2016. Retrieved September 28, 2016.
  2. https://sidbala.com/h-264-is-magic/amp/
  3. "Generation loss". Webopedia. Retrieved September 28, 2016.
  4. Zhang, Michael (February 11, 2015). "Experiment Shows What Happens When You Repost a Photo to Instagram 90 Times". Petapixel. Retrieved September 28, 2016.
  5. Simpson, Isaac (December 4, 2013). "Copying a YouTube video 1,000 times is a descent into hell". The Daily Dot. Retrieved September 28, 2016 – via YouTube.
  6. Zhang, Michael (June 4, 2010). "I Am Sitting in a Room, YouTube Style". Petapixel. Retrieved September 28, 2016.
  7. Higgins, Parker (October 31, 2015). "Mad generation loss". parker higgins dot net. Retrieved September 29, 2016. Mad Generation Loss is a project exploring media encoding and the ways in which imperfect copies can descend into a kind of digital madness. It takes an audio file—here, a recording of Allen Ginsberg reading an excerpt from his seminal poem “Howl”–and adds another layer of mp3 encoding to each second of the sound.
  8. Residuum (January 27, 2005). "audio I Am Sitting in A Room". Retrieved September 29, 2016 – via Internet Archive.
This article is issued from Wikipedia. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.