Compression artifact
|
|
Compression-artifacts.jpg
A compression artifact or artefact is a particular type of data error that is typically the result of quantization in lossy data compression. Where transform coding is used, they are typically in the form of one of the basis functions of the transform coder's transform space.
Compression artifacts occur in many common media such as DVDs and also common computer file formats such as MP3 or MPG files. They are easily seen on many DVDs or MPG video files as described below.
Compression artifacts are a result of the latent error in lossy data compression. Non-compressed (such as Laserdiscs, Audio CDs, and WAV) or lossless compressed (FLAC, PNG, etc.) media, do not suffer from compression artifacts.
| Contents |
Compression artifacts in picture coding
Where the DCT image transform is used, for example, they are often 8x8 pixel squares, containing a stripe or "checkerboard" pattern.
Where predictive coding of motion pictures is used, as in MPEG-1, compression artifacts tend to remain on several generations of decompressed frames, leading to a "painting" effect being seen, as if the picture were being painted by an unseen artist's paint-brush.
Where motion prediction is used, as in MPEG-2 or MPEG-4, compression artifacts tend to move with the optic flow of the image, leading to a peculiar effect, part way between a painting effect and "grime" that moves with objects in the scene.
Errors in the bit-stream can lead to errors similar to large quantization errors, or can disrupt the parsing of the data stream entirely for a short time, leading to "break-up" of the picture. Where gross errors have occurred in the bit-stream, it is not unknown for decoders to continue to apply 'painting' updates to the damaged picture, creating "ghost image" effects.
To stop the build-up of compression artifacts, most compression systems occasionally send an entire compressed frame without prediction or differencing, usually at the start of a shot and at regular intervals thereafter. In MPEG picture coding, these are known as "I-frames", with the 'I' standing for "intraframe compression."
Compression artifacts in audio coding
One technique is to use a lower bitrate by resampling the audio. By reducing the sample rate, higher frequencies must be removed to conform to the Nyquist-Shannon sampling theorem. If the anti-aliasing filter works imperfectly, digital distortion or aliasing will be heard in the form of inharmonic frequencies reflected around the Nyquist frequency. (e.g. a 22.85 kHz tone processed with a Nyquist frequency of 22.05 kHz will result in a tone of 22.05 - (22.85 - 22.05) = 21.25 kHz. This can be generalised to outputF = NF x 2 - inputF) This may be subtle, but more severe levels of distortion can sound similar to ring modulation. (see Aliasing#An_audio_example) Lowering the amount of data (bits) captured per sample can result in loss of detail and dynamic range in the audio. The loss of quality in both methods will be uniform across the recording.
Another technique is to attempt to remove sounds that typical human hearing cannot perceive. As a human being cannot perceive the difference, the resulting data will be simpler (and thus compress better using lossless techniques). For example, in general human beings are unable to perceive a quiet tone simultaneously with a similar, but louder tone. A lossy compression technique might identify this quiet tone and attempt to remove it. As no algorithm is perfect and tradeoffs can be made to throw away additional data to reduce data rate, this will occasionally lead to perceivable sounds being discarded. As these sounds are, ideally, hard to perceive anyway, the result will generally be of flattening complexity, or muddying the sound.
Many systems attempt to replace the series of samples of audio with other representations. Typically these representations make it easier to attempt to eliminate non-perceivable sounds and make it easier to compress the data using traditional lossless techniques. One common technique is to represent the audio as the sum of a series of sine waves. The representation may not be perfect; in exchange for a more easily compressed description, accuracy may be sacrificed.
Many audio compression systems endeavor to match a target data rate, typically expressed in bits of data per second of audio. When using a constant data rate, simple portions of the recording (a simple tone or silence) will be easily compressed to the target rate; the resulting playback will be highly similar to the source audio. As more complex sections are recorded, the system will be forced to seriously reduce quality to meet the target rate; the resulting playback will display more artifacts. Many audio compression systems support Variable Bit Rate encoding, varying the target rate in an attempt to maintain a constant quality of reproduction.
See also
External links
- ff123's Audio Artifact Training Page (http://ff123.net/training/training.html)de:Artefakt (Technik)