6. Preservation-quality images: enhancement

Enhancement options. The consultants introduced the topic of enhancement by enumerating four algorithms that are especially relevant to manuscript and other document images. They recommended that contrast stretching be considered for this project and offered descriptions of the other three. Although some committee members were at first uncomfortable with the term enhancement, suggesting as it does tampering or distortion, most members agreed that the examples of contrast stretching improved legibility with little or no adverse effect.

The consultants argued that reformatting projects should create a record of the processes that were applied when images are created and should report those processes to users. The scanning software written by Picture Elements for this project automatically generates a readable-text log file that includes a detailed record of the scanning process.

Contrast stretching. Contrast stretching in its simplest definition is "making paper lighter and writing darker." In the extreme example presented to the committee, the enhancement rescued a document that was illegible when scanned with the system's default settings. The sample document was a mimeograph print on dark red paper.

The consultants reported that in the production phase of this project, contrast stretching would be accomplished in an automated mode by applying a look-up table to each pixel's brightness value to increase the spacing between brightnesses for those ranges of brightnesses more likely to contain information. This can be termed an enhancement, and while in many cases it is only imperfectly reversible, the nature of the modification can, in principle, be stored with the image data as a form of metadata about the provenance of the image.

Depending on the file format used, it may not be necessary to modify every pixel in performing this operation (which is very slow). Some file formats permit a brightness correction curve to be stored in the file which is only used to brighten the image at display time, often using only the video display's palette (which is very fast, and just as effective). This approach is fully reversible, since a future technician could simply drop the correction curve from the file to remove the enhancement. Since this option does not exist for JPEG-compressed files, it was not considered for this project.

In addition, each scanner has a characteristic way in which its inherent brightness curve is imperfect. With test charts, this can be established and documented. In this context, it is worth noting that avoiding any brightening of an image out of a sense that this increases its fidelity is an imperfect strategy; every scanner is introducing its own brightness distortion. If no hardware support exists within the scanner or its interface electronics for the histogram operation which is needed in this brightening process (as is currently typical), the creation of this data in software can slow down scanning cycle times somewhat, regardless of which brightening approach is used.

There was discussion (but there were no examples) of the use of contrast stretching for what might be considered to be an aesthetic purpose, counteracting an effect sometimes seen in tonal document images where paper may be rendered in a middle gray tone. Some brightening of the paper is a natural side effect of (the recommended) objective contrast enhancement method, differing from a subjective one only in the exact goal of the operation. An objective, repeatable contrast enhancement operation will always improve the aesthetics of the image, but in striving not to lose information, may not whiten the paper to the extent a human operator might choose. To a certain degree, this variant form of contrast stretching can be produced in imaging software by adjusting the brightness and contrast settings.

It is worth noting that contrast stretching after JPEG compression is likely to increase the visibility of whatever JPEG compression artifacts are present. The consultants recommended a procedure (and developed an application program) that performed the contrast stretching before JPEG compression.

Deskewing. This algorithm removes skew caused by poor parallelism of the document's edges to the scanning axes. In a platen-type (flatbed) scanning operation, this should not be necessary if the operator uses due care in placing the document and in not closing the scanner lid too quickly.

This operation would be of critical importance in a high-speed scanner (of the type described later in this report) where originals are hand-placed onto a staging platform (with alignment aids, more or less) and then slid onto a moving vacuum belt. Considerably more skew could be expected in that situation relative to a flatbed scanning operation.

Deskewing can be performed on either grayscale or binary data. Much superior results are achieved when using grayscale data, however, this operation is so computer-intensive that hardware assistance is required to accomplish it in reasonable times. Grayscale deskewing has only recently become available in some scanner hardware. Picture Elements has drafted a white paper on this topic.

Some automated software tools do exist for processing binary images. Deskewing in the binary produces considerably poorer results as compared to grayscale deskewing (when viewed at higher magnifications), since the jagged, stair-stepped jumps seen in scanned straight lines in a skewed binary image are actually increased in number rather than decreased by binary deskewing. While the visual impression is improved at standard viewing distance by the application of binary deskewing, it nonetheless has an impact on the formation of the fine details. For this reason, the consultants did not recommend binary deskewing and no examples were prepared for the committee.

Despeckling of binary images. Despeckling is the operation of removing unwanted small pieces of black (and sometimes white) in the binary image. If the document backgrounds are relatively clean, this might not be necessary. In general, the consultants noted, despeckling is a useful step and many scanners incorporate it natively. For those that do not, a software enhancement post-process can be used to remove speckles. The thresholding technology used by Picture Elements natively includes a despeckler for up to 4 by 4 black pixels and up to 3 by 3 white pixels. Since many of the manuscripts have noisy backgrounds, this enhancement was recommended for use on all binary images.

Display-time smoothing of JPEG-image backgrounds. An enhancement called AC prediction can be performed on JPEG-compressed images. At high compression levels, JPEG images can be prone to "blockiness" in the flat regions of the background where subtle shading changes occur. AC prediction can reduce the perceptibility of the 8x8 blocks in a JPEG image by causing them to match their neighboring blocks better. As with contrast stretching, this can be applied at display time and can create a more pleasing image for the viewer. The consultants' recommendation was to press display-software designers to include this feature in typical viewing software.


Next Section | Previous Section | Contents