Improved Training GUI

Major back end improvements to the behavior of the training user interface.
Improved documentation and instructions.
Made the image white balance the default option for spring sRGB imagery.
Other minor bug fixes for edge cases.

Bugfixes

Fixed an issue with applying source image projection to the classified output. Removed code used for debugging.

Computational Improvements

• Significant improvements to memory usage and computational efficiency.
• Improved image segmentation algorithm
• Added optional white balance function to handle poor quality images.

Tiffs and Blocks

This version includes improvements and simplifications to the processing code. Input images are now read one block at a time, the block is processed completely, then written to the output file. Smaller aerial images will be processed in a single block (as before), while larger satellite images will be split into many blocks. The default block size is 6400x6400 pixels, but is this size is altered such that no block is smaller than 3200 in either dimension.

Output classification images are now written as .tif files, with the same projection and geolocation information as the input dataset.

Removed:
Split option. The need to customize this parameter no longer exists. The ability to chose the desired block size may be added in future updates.
Parallel processing.

Added:
Ability to incorporate image pansharpening and orthorectification of WV images using PGC scripts.

General code cleanup and simplification.

Version 2

Changed the way segment attributes are calculated and migrated some processes to Cython for improved speed. Reduced the amount of image subdivision during processing which should decrease the occurrence of straight line misclassifications (e.g. an object being classified differently in two adjacent image divisions).

v1.0

Version associated with publication in The Cryosphere journal.