General Compression Tips

Choosing the right codec is important, but its not the only factor you should consider when compressing a video. There are many other areas that you can tweak to get better quality or smaller file sizes.

Resolution

The resolution of NTSC video is 720x480, however, you should only export your video at 720x480 when the final product will be recorded onto videotape and played on a TV. This is because TVs use "pixels" (they're not actually pixels, but for simplicity I will call them that) that are slightly rectangular, while the pixels computers use are square. Thus video that looks fine on a television will look slightly stretched on a computer monitor.

To make your video look the correct size when played on a computer, for full resolution export your video at 640x480. A 640x480 video on a computer will have the same aspect ratio as a 720x480 video on a television. (Note: Since DV is intended to be viewed on a television, the resolution of DV codecs is fixed at 720x480).

If small file sizes are a must, consider reducing your resolution to 320x240 (or lower). Obviously, you'll want to reduce each dimension by the same amount to keep the proper aspect ratio. A few codecs (such as MPEG-1 and H.263) were designed for lower resolutions and actually have better quality at these resolutions. For example, NTSC MPEG-1 is designed for 352x240. When maximized to full screen on a computer monitor, a 352x240 MPEG-1 file will actually look better than the same video encoded at 640x480, even though the smaller file is being blown up more. (My codec guide will tell you if a codec works best with a particular resolution).

Pixel Aspect Ratio

This is related to the difference in resolution between TVs and computers that I explained above. If a video is destined to be on videotape and viewed on a TV, the pixel aspect ratio should be 0.9 (labeled "D1/DV NTSC (0.9)" in Premiere). For video that is to be viewed on a computer, the pixels should be square, or have a aspect ratio of 1.0 ("Square Pixels (1.0)" in Premiere). Usually only video editing programs will give you the option to select the pixel aspect ratio; others will default to square pixels.

Frame Rate

The native frame rate of NTSC is 29.97 fps (it is often casually referred to as 30 fps, but to be correct, always select 29.97 when working with NTSC video). However, if your video is not action intensive, cutting the frame rate can lead to a much smaller file (obviously, if you are demoing a real-time method or something that needs smooth motion, you'll want to keep the full frame rate). There are only a few certain frame rates for NTSC that will yield acceptable results. These are: 15, 10, 7.5, 6, and 5 fps. If you want to reduce the frame rate, stick to one of these values.

Cropping

Cropping is a good way to reduce the size of the video while also getting rid of unnecessary or unwanted elements. When played on a television, NTSC video is overscanned, meaning that the far edges of the video are actually not seen. However, these edges are seen when the same video is viewed on a computer. Since this area usually does not contain any needed information (since it wouldn't be seen on a television), it can usually be safely cut off without losing anything. In fact, you might find that unwanted elements are visible in this area, such as menu bars from a computer demo or things you thought weren't in the shot. In these cases, you'll probably definitely want to crop the video. For 720x480 NTSC video, the overscan area that can be cut off is 36 pixels from the left and right and 24 pixels from the top and bottom.

Another time when you might want to crop the video is if the video is letterboxed. There's no point in wasting your bits on those black bars, so with cropping you can easily lop them off and save disk space without losing anything.

Scaling

If you are scaling your video, you are sometimes presented with an option as to which scaling algorithm the program uses. The differences between most scaling algorithms are subtle, with the exception of Nearest Neighbor, which you should avoid like the plague. The methods are:

Bicubic- The most common method, found in Adobe Premiere and After Effects. Bicubic is good for scaling for as much as twice the size of the original and down to half the size of the original. Anything more or less and blockiness becomes rather noticeable.

Sine- Less common that bicubic. Use sine (if available) for when you greatly want to enlarge or reduce your video (more than 2x), for it performs better than bicubic in extreme cases.

Bilinear- Samples pixels adjancent to the source pixel to determine the output. Old and certainly not the best, but better than nearest neighbor.

Nearest Neighbor- As stated above, the worst scaling algorithm. It simply duplicates the nearest pixel, yielding a very blocky result.

Deinterlacing

NTSC's interlaced nature looks fine when displayed on interlaced TVs, yet on progressive scanning computer monitors interlace artifacts can occur (see Interlacing in NTSC Basics for more info). Interlace artifacts appear as thin vertical lines in a comb- like pattern that appear in areas of the screen where there is movement (sometimes called "feathering"). Deinterlacing eliminates these artifacts, so it is a good idea to deinterlace any videos that originated from an interlaced source (e.g. non-progressive camera footage, scan converter footage).

Things are complicated by the fact that not all deinterlacers are created equal. There are various methods for deinterlacing, some creating much better results than others. Some programs give you a choice as to which deinterlacing method to use. Here are some of the most common (the names can vary):

Weave- This method simply combines the two fields into one progressive frame, which leads to the feathering artifacts described above and is thus not good for high-motion video. Weave is good for sources that were originally progressive (such as film), converted to an interlaced source (TV, DVD) and that you now want progressive again.

Bob- With bob, one field is discarded and the remaining field is doubled to create a complete frame. For the next frame, the process is repeated with the second field (essentially the frame rate is doubled). This eliminates any feathering caused with weave, but the result is that you're throwing away half your resolution, so non-moving objects can look worse. It is so named because horizontal lines (like in text) in the video can appear to "bob" up and down. Good for a source that has always been interlaced (e.g. TV).

Blend- The two fields are blurred together to make the interlace artifacts less noticeable. This causes a halo or ghost effect in motion areas. Because of this, this method isn't used very often.

Smart (or Adaptive)- This method analyzes the video to find areas that are in motion. For areas that are not moving, a simple weave is performed. For moving areas, either one field's information is thrown away and new data is interpolated or the fields are blended (most smart deinterlacers will let you choose which method want). Deinterlacers of this kind also usually allow you to set a threshold that determines how much movement is required before the deinterlacing will begin. This method has the advantage of only deinterlacing where it is needed (in the moving areas) and preserving resolution elsewhere.

My personal favorite is to use the Smart Deinterlacer filter (set to interpolate) included with VirtualDub. Tmpgenc also has an excellent deinterlacer if you are making mpegs. VideoMach's deinterlacer is pretty good, while Premiere's only performs a simple weave and is pretty much useless in eliminating interlace artifacts.

Audio

When trying to get small file sizes, people often obsess over tweaking the video yet completely forget about audio. Uncompressed audio takes up far less space than uncompressed video, yet it still takes up around 10 MB per minute, so to get a movie that has audio to a really small size some audio processing must occur.

There are various audio compression codecs available, but most of the time you can dramatically reduce the audio bandwidth by adjusting other audio parameters. A soundtrack that is all speech can really be squeezed into a small file, since intelligibility is usually the only concern.

Sampling Rate: The default sampling rate for our DV devices is 48 kHz, which is overkill for the web, especially when it's just speech (since the sampling rate for CDs is 44.1 kHz, DV audio is actually better than CD quality). Reducing the rate to 22.5 kHz will produce results that are fine for most. For just speech, even 11.025 kHz works, although your narrator will sound like he is on AM radio.

Sampling Resolution: This is either 16-bit or 8-bit. 8-bit resolution has a worse signal- to-noise ratio and a narrower dynamic range (which means it's harder to hear quiet sounds). If you all you have is speech and you really don't care about the quality (as long as it is intelligible), then you might be able to get away with 8-bit. However, your audio may be a bit noisy. 16-bit should be certainly used if you have music in your movie, and some audio codecs only work in 16-bit.

Mono vs. Stereo: I can't tell you how many times I see people whose videos consist of just a voice-over and yet encode in stereo. Unless you are using a stereo microphone (chances are you're not) or are recording with two separate microphones, that voice-over you recorded is in mono. When you encode your project's audio in stereo, all that happens is that the mono track is duplicating and thus your audio takes up twice as much space as is necessary. Besides, why would you even need to hear a voice-over in stereo? Bottom line: when all you have is a voice-over, go mono.

Return to Video Guide