Experiments+in+recreation+of+line-structure+through++block-matching

An experiment was conducted to see if it may be possible to use block-matching techniques to detect and quantify the geometrical distortion introduced by the film recording process. If it were possible then reversal of the spatial distortion would be possible. The possible advantages of this would be:

i) Recreation of the line structure may enable recreation of the original PAL signal, and decoding of this would then solve colour recovery. ii) A technique for automatic spatial repositioning of the luminance from the film recording could replace the traditional method of using Charisma for image alignment in the reapplication of colour from a poor quality off-air recording. iii) Recreation of line structure would made it possible to separate the two video fields on the film recording from one another. The 'peeling apart' of the two video fields would mean that temporal information is recovered, and the 'video look' can be restored to the material. This would be an alternative to use of the VidFire technique.


 * The Experiment**

A section from a SD colour clip of Top of the Pops was ingested as a raw UYVY file. The matching section of the same programme from an HD transfer of a 16 mm film recording negative was also ingested as a raw UYVY file.


 * A C++ program was written to take in a SD (720x576) video squence and an HD (1920x1080) video sequence.
 * Each SD frame was divided into small blocks.
 * An attempt was made to match each block with a similar-looking region of the HD image. (Firstly, scaling up block size to match that expected of the HD image).
 * The search range within the HD image was limited to a relatively confined area - based upon expectations of where a match was likely to occur.
 * A map was created to record the horizontal and vertical offset between each SD block and that of the best match within the HD image.

For illustrative purposes, a new SD image was created such that sections of the HD image were mapped back to the correct SD positions identified. (In practice, a technique such as [|mesh-warping] would be best suited to this).

Frames from this experiment are shown below:


 * [[image:http://www.insell.co.uk/colourisation/ref.jpg]]©BBC || **SD video reference image** ||
 * [[image:http://www.insell.co.uk/colourisation/source.jpg]]©BBC || **The same frame taken from the film recording** ||
 * [[image:http://www.insell.co.uk/colourisation/undistor.jpg]]©BBC || **The 'undistorted' resultant image** ||

-- James Insell - 13/03/2007