Subsampling Options
Options for the subsampling process that happens after the extraction of the selector mask.
Decreases the number of pixels in the image's foreground layer. Because foreground objects vary little in color, you can subsample this layer without affecting their appearance. The foreground of a 300 dpi image, therefore, can be subsampled to 25 dpi, which corresponds to a value of 12 (300 dpi / 25 dpi = 12). Assigning a value of 12 to the fg-subsample option usually ensures that each letter in a word is the correct color.
Option type
Pixel size
Range
3..12
Default
12
Command line
yes
Decreases the number of pixels in the image's background layer. Because background objects are generally low-contrast and smooth, you can subsample this layer without affecting their appearance. The background of a 300 dpi image, therefore, can be subsampled to 100 dpi, which corresponds to a value of 3 (300 dpi / 100 dpi = 3). Assigning a value of 3 to the bg-subsample option decreases the size of the output file and ensures that the letters or parts of letters that were not placed in the foreground are still readable.
Because background objects tend to appear blurry below 100 dpi, values 1 and 2 should be used only in the following instances:
The segmenter places all objects in the appropriate layers and the
background does not contain varying colors (for example, colored paper)
The original resolution of the scanned image is below 100 dpi
Option type
Pixel size
Range
1..6
1
300dpi bg
6
50dpi bg
Default
3
Command line
Yes
Increases the number of pixels in an image's mask layer. Because mask objects (text and line drawings) have sharp edges, they must be encoded at a high resolution to maintain their contrast and readability. Documents, therefore, should be scanned at 300 dpi so that the mask layer can be encoded at the recommended 300 dpi. If you are encoding an image with a lower resolution, for example, 100 dpi, use this option to upsample the image's mask layer by a value of 3 (3 * 100 dpi = 300 dpi). This new dpi is then delivered to the segmenter during the encoding process.
Option type
Pixel size
Range
1..3
Default
1
Command line
Yes
Decreases (or maintains) the size of the image for segmentation purposes. The Document Express segmenter is designed to be most effective with images that are 300 dpi. If you are encoding an image that is over 400 dpi, assign a value of 2 to the resolution-multiplier option. Doing so alerts the segmenter that it should skip every other pixel and row when it analyzes the image, thereby reducing the image's scanned resolution by half. If you are encoding an image that is below 400 dpi, assign a value of 1 to this option to maintain the current resolution.
In theory, resolution-multiplier=(image-dpi/400)+1.
Option type
Pixel size
Range
1,2
1
Less than 400dpi
2
More than 400dpi
Default
1
Command line
Yes
Expert
The original luminance image and, as a
consequence, the mask, have a resolution approximately resolution-multiplier
times 300dpi.
The default value is 1.
When it is set to two, shall all the other sizes be multiplied by 2?
In practice, no, as this implies that we process a 600 dpi image as if it had the same quality as a 300 dpi image. This is not true: 600 dpi edges are much sharper. Moreover, if you use a 600 dpi image as input, you should also expect a higher quality DjVu® output.
NOTE: Choosing a resolution-multiplier value greater than 1 has an impact on adaptive thresholding.