DV Pix - Image Artifacts
Pix: Sampling
>Pix: Artifacts<
Pix: Defects
 Pix: Generations
Pix: Codecs
 
Text: The DV Format
Mosquito Noise: full frame
DV title frame, white text on blue: full screen

Mosquito Noise: detail
DV title frame, white text on blue: 72x48 pixel detail
both images (c) 1997 Texas Instruments, Origin8 Video

Mosquito Noise (a.k.a. feathering, scatter, "critters", "gribblies") results from high-frequency data lost in the compression/decompression process. It's always within 8 pixels of a hard-to-compress edge or detail, and a common artifact in any DCT based compressions such as DV, JPEG, or MPEG. Unfiltered titles (here, from Premiere 4.2's title generator) on plain blue backgrounds make the clearest demo images, but these artifacts are also frequently found in DV pix of dense clusters of leaves, sharp diagonal details, and the like, especially where the background image is itself quiescent and bright enough so that the noise isn't lost in the shadows. 

In this 72-pixel by 48-pixel detail, notice the noisy four rows of pixels above and adjacent to "1997" and the general lack of uniformity between the left edge of the image and the "1".

Quilting: full frame
Quilting demo, full screen
(note: low-quality JPEG image not representative of original's quality)

Quilting: detail
Quilting demo, 72x48 pixel detail
 both images (c) 1998 Origin8 Video

Quilting shows up as discontinuities between adjacent DCT blocks in an image. They are most noticeable in DV on straight diagonal lines that are slightly tilted from the horizontal or vertical. Quilting is also common to other DCT compressors like those used in JPEG and MPEG. 

Quilting is most often seen on DV footage as a "fixed pattern" distortion during slow pans or tilts. It can be accentuated by maximizing the "sharpness" or "aperture" control on a high-resolution video monitor (such as the 600-line Sony PVM series), which really brings it out, often to a frightening degree. 

The 72-pixel by 48-pixel detail of the larger scene shows quilting on the edge of the table. The small arrows at the bottom of the image are aligned with the DCT block boundaries, every 8 pixels across the image. 

??? Motion Blocking, where detail diminishes in areas of fast motion (causing the DV codec to compress the two fields separately), does exist --  but I've not been able to find a still or series of stills that illustrate it with detail and clarity. 

Technical Details: images were shot on a Sony DCR-VX1000 3-chip DV camcorder. Titles were rendered in Adobe Premiere 4.2 and written to tape through the Adaptec DVsoft codec supplied with DPS Spark. The tapes were played in a DHR-1000 DV VTR and captured digitally using the DPS Spark (Adaptec AHA-8940) IEEE-1394 I/O card on a P133 Wintel machine. The resulting AVI video clips were opened in Adobe Premiere 4.2 and the desired frames were exported as 24-bit Windows BMP files using the Adaptec DVSoft codec; the BMPs were then read into Adobe Photoshop LE. Full-screen images were downsized to 320x240 using bicubic interpolation and saved as low-quality JPEGs for efficient transmission; full screen images are shown as references only, not as samples of the quality inherent in the original BMPs. Detail images are upsampled from the original BMPs using "nearest neighbor" interpolation and saved as high-quality JPEGs to preserve visual quality; these JPEGs are very close in appearance to the original images. Images are shown with square pixels, though for a proper 4:3 aspect ratio the pixels should be 12.5% taller than wide.

Copyright (c) 1998 by Adam J. Wilt.
You are granted a nonexclusive right to duplicate, print, link to, frame, or otherwise repurpose this material,
as long as all authorship, ownership and copyright information is preserved, and a link to this site is displayed.


Contact me via email

 
 
 
 

Last updated 19 June 1999.