GSoC 2015 - Cloud removal from satellite imagery for GeoTrellis

GeoTrellis is an open-source, scala-based geospatial data processing framework. Clouds and cloud shadows often obscure parts of images acquired by satellites or other space-borne sensors. However, useful image data could be extracted from other relatively non-cloudy images of the same geographical location that can be used to ‘fill’ these obscure parts. When combined together with the right algorithms, these images can offer a cloud-free (or nearly cloud-free) view of that area.

This project aimed to enable GeoTrellis to remove clouds from a large set of imagery. Combined with it’s power to process large amount of raster data, it would enable users to run this process over any amount of imagery, in a scalable way. Scala was a completely new language for me before the start of the project, but my mentor Rob, was a great guide throughout the project.

We tested our algorithms on this geographical location as it’s cloudy and snowy here most part of the year.

Some input images:

image1 image2 image3 image4 image5 image6

Resultant image:


comments powered by Disqus