16.36. image_subset

Given a set of overlapping georeferenced images, this program will extract a subset that results in almost the same coverage as the original images.

This program is slow, as the algorithm complexity is proportional to the square of number of input images and the number of output pixels. It is best used with at most 100-200 images, each of dimensions of 1000 - 2000 pixels.

16.36.1. Overview

A threshold determines if a pixel is considered to add to the coverage.

This program finds the image that contributes the most pixels no less than the threshold, then the image that contributes the most additional pixels, etc.

The produced subset is saved to an output list, with the image name and number of contributing pixels on each line. The images are sorted in decreasing order of contribution to coverage.

16.36.2. Background and example

This program was developed as an auxiliary tool for Shape-from-shading processing (Section 11.1). For that, it is desired to have a very large number of images of diverse illumination to be able to coregister them all. However, once that is done, just a representative subset of images is needed for SfS, as using the full set can be prohibitive.

The following way of invoking this tool is suggested. First, break up the input image set into several groups, by Sun azimuth (Section 11.10.4), with 50 - 150 images in each group.

For the images in each group, create a list of mapprojected images at a low resolution, for example, at 1/16 of original image resolution. This can be accomplished by using sub images produced by stereo_gui (Section 16.70), whose resolution can be further reduced with a command such as:

gdalwarp -r cubicspline -tr <gridx> <gridy> input.tif output.tif

Then, for each group, this program can be called as:

image_subset            \
  --threshold 0.01      \
  --image-list list.txt \
  -o subset.txt

Note, as before, that this program can take many hours. A progress bar is displayed and helps track its advancement.

For a very large terrain, it is suggested to break it up into quadrants with some overlap, collect all relevant images for each quadrant, and then process these quadrants separately. This will result in a faster runtime and produce images that are more relevant to each smaller area.

A good threshold can be found by clicking on pixels of representative images in stereo_gui and observing the pixel values printed in the terminal.

16.36.3. Validation

When this program finishes, overlay the images in the produced list and inspect if they cover the desired area in decreasing order of contribution to coverage.

The program dem_mosaic (Section 16.20) with the --max option can be helpful in determining if the produced subset coverage is about the same as for the original images.

16.36.4. Command-line options

--image-list <string (default: “”)>

The list of input images.

--output-list, -o <string (default: “”)>

The file having the produced image subset, and for each image the number of contributing pixels (sorted in decreasing order of contribution).

--threshold <double (default: NaN)>

The image threshold. Pixels no less than this will contribute to the coverage.

--t_projwin <float float float float (default: 0 0 0 0)>

Specify a custom extent in projected coordinates in which to evaluate the coverage. The format is min_x min_y max_x max_y or min_x max_y max_x min_y, with no quotes. In this mode all input images must use the same projection.

-v, --version

Display the version of software.

-h, --help

Display this help message.