Insect models of illumination-invariant skyline extraction from UV and green channels. (7th September 2015)

• Record Type: Journal Article
• Authors: Differt, Dario; Möller, Ralf
• Is Part Of: Journal of theoretical biology. Volume 380(2015)
• Page Start: 444
• Page End: 462
• Publication Date: 2015-09-07
• Subjects: Insect vision -- Color vision -- Linear separation; Biology -- Periodicals; Biological Science Disciplines -- Periodicals; Biology -- Periodicals; Biologie -- Périodiques; Theoretische biologie; Biology; Periodicals; 571.05
• DOI: 10.1016/j.jtbi.2015.06.020
• Languages: English
• ISSNs: 0022-5193

Abstract

Abstract: Experiments have shown that the skyline is an important visual cue for navigating insects. However, the comparison between two snapshots collected at different times of day is a complex task due to possible illumination changes. In this study we examine whether the information from two different color channels (UV and green, which are also available for many insects) can be used to obtain an illumination-invariant separation between the sky and ground. We collected UV and green images of seven different scenes over entire days, in which natural and artificial objects are visible in front of the sky. With the collected data we want to find answers to the following two questions: 'Does UV/green contrast vision increase the quality of separation compared to UV-only vision?' and 'What yields a better performance: separation methods based on a fixed threshold (global separation techniques) or separation methods which adapt the threshold dependent on the input image (local separation techniques)?' We implemented several linear separation techniques and found that UV/green contrast only marginally increases the quality of global separation in comparison to UV-only, and that local separation techniques are superior to global separation techniques. Abstract : Highlights: We collected multi-spectral images (UV/green) of panoramas containing vegetation. We examined linear classification methods on UV/green data between sky and ground objects. Online learning techniques generally exceed offline techniques. UV-only separation does generally not impair the classification compared to UV/green.