When scanning transparencies, and to a lesser extent reflection copy, images it is common that unwanted artifacts such as dust and scratches are included in the scanned image.� To reduce this effect great efforts in cleanliness need to be applied. Unfortunately this is not enough and it is impossible to avoid dust on transparencies. Scratches of course are impossible to remove once they have occurred.� Traditionally this has been resolved with the use of oil mounting which uses a refractive index matched fluid to “fill in” the scratches. Unfortunately this is a skilled job which only high end users are prepared to perform but all users do not want scratches and dust on their transparencies to show up on the scanned image.

To resolve this software correction of the scanned image can be performed which will remove the dust and scratches without removing the image detail.� One alternative to this software correction is to use additional scans with different lighting arrangements. This will illuminate the three-dimensional objects such as scratches and dust in a different way, which can then be used to discriminate the objects from the image detail. Another alternative is to use a second scan at a different wavelength which is transparent to the image detail, the photographic dyes, but is still effected by the dust and scratches.

This report describes:

1.       A software only dust and scratch detection system and the subsequent image processing to remove the dust and scratches from the scanned image.

2.       A method of processing a second scan to identify dust and scratches and the subsequent image processing to remove the dust and scratches from the scanned image.

3.       A method of combining both above processes.

The overall process for the software only scratch and dust removal system is to:

1.       Identify dust and scratches from their high frequency content in four directions (0, 45, 90 and 135 degrees).

2.       Check that the high frequency items are surrounded by uniform background in their respective directions. If they are mark pixel as dust.

3.       If the high frequency content is very strong and the background is uniform then mark all surrounding pixels as dust.

4.       Check that at least two directions identify the pixel as dust. If they do then the pixel is dust. If they do not the pixel is not dust.

5.       For every pixel that is marked as dust perform a linear interpolation from the nearest two pixels which are not marked as dust.

For each stage the more detail description of the operation and reasoning for the stage is as follows:

Identify dust and scratches from their high frequency content in four directions (0, 45, 90 and 135 degrees).

Dust and scratches are normally actually difficult to see in raw unprocessed scanned images.� It is only when image sharpening is applied that they become easily visible. This...