Open AccessReference data preparation for complex satellite image segmentation
Author(s) -
Sharma Ashu,
Ghosh Jayanta Kumar,
Kolay Saptarshi
Publication year - 2020
Publication title -
iet image processing
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.401
H-Index - 45
eISSN - 1751-9667
pISSN - 1751-9659
DOI - 10.1049/iet-ipr.2019.0234
Subject(s) - segmentation , computer science , artificial intelligence , computer vision , gestalt psychology , pattern recognition (psychology) , scale space segmentation , reference data , ambiguity , image segmentation , perception , data mining , neuroscience , biology , programming language
The high spatial resolution satellite images covering multiple objects of the urban area are visually complex in nature. This visual complexity causes ambiguity during segmentation of such images when the targets are unknown. In such case, reference data are required to assess the segmentation methods. Due to excellence of humans in visual analysis, the presented work has attempted for a psycho‐visual approach to prepare the reference segmented complex HRS images. The reference data have has prepared through the correlation of the quantified eye‐tracking data (metrics) and corresponding concurrent think‐aloud (CTA) data for each of the created segments. Segments get updated based on Gestalt principles and Gibson's theory. Those segments having the best correlation between metrics and CTA data have been opted as the final output. The results suggest that the functional grouping of objects while preferring perceptual grouping for segments drawing conforms the most to the participants' verbal response. The final results have been compared with the existing notion of full segmentation used for complex images. The comparison also proffers the superiority of the proposed segmentation over full segmentation to be used as reference. In the future, a large number of images may be used to prepare better reference data.