4. Digital Video Basics 4:2:2 sampling 4:2:0 sampling 4:1:1 sampling 4:2:2 – High End DV (Digital Betacam, DVCPro50) 4:2:0 – MPEG 1 & 2 4:1:1 – DV and DVCAM
5.
6.
7.
8.
9.
10. One River Media Codec Test Image Images courtesy of Marco Solorio, One River Media
11. Black Magic 8-Bit 4:2:2 Uncompressed Codec Images courtesy of Marco Solorio, One River Media
12. Black Magic 8-Bit: 10 th Generation Images courtesy of Marco Solorio, One River Media
13. One River Codec Test Images courtesy of Marco Solorio, One River Media
Comments courtesy of Marco Solorio, One River Media: NTSC Color Bars - These colors were taken from Final Cut Pro 3. It's nice to see how well these codecs keep their color integrity to the time old standard color bars. Also look between the color bars and you'll notice edges combining and smearing colors together in some codecs. Parking Meter - This somewhat represents what a true-life camera shot would look like. The top of the handle also makes for a good shadow test. In some codecs, detail is lost in most of the parking meter image. Chroma Text - Under most situations, a compressed color space of 4:2:2 will yield less than desirable results with a test like this. The text simply can't hold up to the chroma value of the bluish background color. Luma Text - More often than not, a luminance test like this will favor well. Only poor codecs will fail this test. Color Filtering - Ah, one of the best tests by far. This shows how much the codec is color filtering the image. Some codecs lose this test altogether by changing part of the red aliased line to gray. The key is to enlarge the image after you render it so you can really see the effect of color filtering. The dynamic image selection below shows the different codec's color filtering side by side. Note that the image has been enlarged 300% using "Nearest Neighbor" instead of the default "Bicubic" so Photoshop does not interpolate new, colored pixels. Thanks to Matt Silverman for leading me on to this little, but powerful test. Gradient Banding - This is where all of this started! It was initially due to banding that made me first look into testing different uncompressed codecs. Back in my Media 100 days, banding was something you had to deal with. But with a new NLE purchase in mind, I wanted to see which uncompressed NLE did away with this age-old problem. For this test, I created a solid layer in Adobe After Effects. I added a Mandelbrot color fractal and blurred it out with a value of 120. The colors slowly change over time to give codecs like MPEG and Cinepak a tough time with spatial interpolation. As for the gradient, some codecs just can't handle it and banding is a result. RGB Bars - To get a good effect of color filtering (or lack there of) combined with the color space limitation of YUV video, pure red, green and blue colors are used and placed at a slant. If the bars were perfectly straight, then the effect may not produce large enough blocking. Some codecs have larger blocking in this area than others, even in the same color space. RGB Bull's Eye - Pure RGB circles are stacked upon each other to see how much the colors smear into each other. In most cases, the blue and green circles stack up okay, but the red circle usually becomes aliased. Single Pixel Lines - A test to see how well a codec can accurately encode single pixel lines of differing colors. Notice also that the green line is slightly off the background green color. Makes for interesting results. Compared to hair, these lines are thick. Color Grid - Using a grid of lines that are both 1-pixel and 2-pixel in width, we can see how much color smear a codec may produce. This test can also show the relationship between the grid's lines and a codecs compression block structure (like a highly compressed Motion JPEG render). Random Noise - The black and white noise test will typically fair well (sometimes perfect) for the 4:2:2 uncompressed bunch. The color noise test however can't replicate perfectly even by the best 4:2:2 uncompressed codec... the colors simply smear out due to the compressed color space limitation. Interlace - This test represents the effect of compression interacting with interlaced fields. If there is a lot of compression or color space artifacts, then this can result in messy interlace lines, which could result in flickery playback. RGB Circles - Much like the "Bull's Eye" test, this test shows how color space blocking can affect an RGB area that is not constant (unlike the RGB bars that are constant). Simple Color Spectrum - This spectrum is a little like the other color spectrum, except that this spectrum has 30-pixel high lines that should stay constant. If a codec contours, then what should be a 30-pixel high color will become many colors. Grayscale Ramp - Exactly like the Simple Color Spectrum, but with monochromatic values. A good codec will replicate all the pixels in this ramp. Some codecs however, induce coloring in these pixels. HSL Spectrum - Although there isn't enough area in this image to display all the colors in an HSL spectrum, it does give us a good idea of any color changes or contouring. 16-bit Ramp - This test is really only useful for 10-bit or higher codecs. To the far left of the red ramp is a 16-bit RGB value of [32768,0,0] and to the far right of the ramp is a 16-bit RGB value of [32768,257,257]. In 8-bit fashion, this is only a difference of a couple RGB values, whereas in 16-bit mode, it's a difference of a few hundred RGB values. Seeing how well a codec can represent these microscopic changes is a true test to its replication accuracy.
For uncompressed case, we assume 141.5 GB for video + another 1.5 GB for 48 KHz 24-bit stereo audio for a single master MPEG4 streams are assumed to be 5 mbps and 0.5 mbps.
http://www.aiim.org/documents/costmagstorage.pdf Assoc. for Information & Image Management 10-year projection of disk prices