Citizen Science-based Labeling of Imprecisely Segmented Images: Case Study and Preliminary Results

One of the most important objectives of remote sensing image processing is the identification of objects on an image. Automatic and correct object identification is hard to achieve, since the intermediate steps (including segmentation) may yield imprecise results. After segmentation, the image segments must be labeled, i.e., their classes must be identified. The approaches to this are either manual, requiring extensive and manual annotation by a trained user, which is by definition expensive, or automatic or semi-automatic but may incur in errors caused by the segmentation step and/or lack of high level knowledge used in segment labeling. We consider another approach: using volunteers and collective intelligence to label the regions accordingly to a minimum set of rules. The objective of this paper is to use the concepts of citizen science in labeling segments from satellite images for urban areas. This paper shows whether citizen science may improve the process of interpretation of scenes or labeling polygons (regions) extracted from urban satellite image through segmentation process and also presents some preliminary results.