Within the last year or so I remember reading a description of some new software products and algorithms involving video streams. That would include video tape or even filmed movies.
I didn't see a lot of details, for the reason that the providers of the software were protecting their intellectual property.
I think the principle is based on the fact that successive frames of video contain the same "information" but are impacted by "noise" sources in different ways and to different extents.
Let's imagine that we have a vido of someone moving in front of a store and that there are signs in the windows of the store. Imagine that a particular sign consists of black letters on a white background. A single frame may show a blurred unreadable area where the printed information is. Frames later on will show a similar blur, but the blur will not be identical.
There are many reasons why the blurred sign is unreadable on a single frame. The graininess of the recording. Slight vibration of the camera. Imperfections in the camera lens. Electical noise in the circuitry of a recording device. Even noise in the playback system can contribute to inability to read the sign.
The ability to "enhance" the image of the sign depends upon the fact that the noise part of the video signal is unrelated to the information on the sign.
At a spot in the image which is supposed to be black, the noise might make the spot white ( or just less black ). Contrarily, at another spot in the image which is supposed to be white, the noise might make it black or less white.
If we have available a series of video frames which contain a blurred image of the same black letters on a white background, the noise effects can be averaged out.
For each frame, a determination will be made of where the center of the blur is both vertically and horizontally. This process of locating the center will be less affected by noise because there is a greater area of the image involved in the determination.
Once the center of the blur has been located in each frame of the video, the position of each blur is adjusted so as to put them all at the same position for further enhancing.
Now each point in every blur should represent the same point in the original image of the writing on the sign. Each point should be either white or black in all the images, but each image will have some noise effect. In order to make a determination of whether some point in the image should be white or black in the enhanced image, various kinds of "averaging" can be used. One type of "averaging" is just that, an average of the intensity at the same point in every frame. Other types of averaging can be applied. Which types to apply may depend on what assumptions can be made about the original information.
If we know that we are looking at a vertical edge of a letter, then we can make assumptions which help to determine just where that edge really is.
I recall from the original description of this capability that it could create still photograph quality images of faces from a stream of standard video tape images.