Source Image
This image shows a single frame of video from the film Gladiator (Don't worry I'm not making copies, this is just for the demonstration). As you can see there are really two images in this frame, and if played back on a computer this frame would be displayed as shown below, this generally looks dreadful if there is any movement! (and video is a but pointless without movement)
In reality on a TV these two images would be displayed independently, this is the technique know as interlacing, a method for effectively doubling the frame rate of a video but also maintaining the full resolution if there is no motion. So the trick to converting video to compress, play back and edit on computers (also on DVDs and VCDs) is separate out these different fields, ideally without loosing the full resolution of the image.
Zoomed in image of the above
Field Drop
By dropping one of the two fields and replacing it with a copy of the other field you can easily get an image with none of the interlacing effect in it. This is by far the fastest method, but it does loose the full resolution, and this drop in quality is rarely acceptable (but I admit hardly perceptible in this image :)
Because only half the data stored is being used the amount of noise in the image appears to be doubled, perhaps one of you mathematicians out there can explain why?
Blend
The blend technique combines both images on all lines of the image, removing the interlace lines, but instead causing a double image effect. This keeps the full resolution but the double image is easily noticed, especially with fast moving objects.
Smart / Area based
Smart and area based de-interlacers use a combination of the two techniques above, they use the blend method, but on pixels where motion is detected, i.e. where the two images do not appear the same, the field drop method is used. With the frame used here everything in the field is moving.. the crowd in the background is moving right and Maximum (the hero of the film) is moving left, if the smart or area based filters work correctly they should just use the field drop method for the whole image. When the camera is held still these filters are perfect, unfortunately in video this is only sometimes the case.
Steady-De-Interlace (Stage 1)
The Steady-De-Interlace plug-in really expands on the Smart or Area based filters. The problem with these filters is that unless the camera is being held perfectly still the blending does not occur correctly and far too much of the image has to be based on just a single field as everything (i.e. the camera) is moving. To solve this problem Steady-De-Interlace works in two stages, the first stage aligns the two fields so that the majority of the image, usually the background, is held steady between the two fields. The second stage is basically the same as smart and area based filters but is far more effective as the majority of the two fields are aligned and can be blended.
After stage one the majority of the two frames are aligned (this is a conceptual stage, an image like this is never actually produced)
Where both fields match the resulting image is of the best resolution possible, this is noticeably higher then the blend filter can produce, especially with high contrast edges at an angle closer to the horizontal then the vertical.
Steady-De-Interlace (Stage 2)
Like the smart filters, where the fields don't match only one field is shown. As you can see the only problem with this technique is as the fields don't exactly overlap there is an area on the left which is not covered by one of the fields, this is solved by dynamically zooming the image, this zoom will be hardly perceived by the viewer.
note 1: All of these images are simulations produced in Paint Shop Pro, and so are only representations of the different techniques.
note 2: I make no claims of ownership and no copyright infringement is intended