Computer vision vs human perception in remote sensing image analysis: time to move on

Arianna Traviglia, Karsten Lambers

Abstract


The (slow) emergence of semi-automated or supervised detection techniques to identify anthropogenic features over remote sensing imagery have received mixed reception in the past decade, with critics stressing the superiority of human vision and the irreplaceability of human judgement in recognising archaeological features, and supporters working toward the development of (semi)automated computer vision methodologies to streamline the screening of aerial/satellite imagery. This limited development has been due to a number of reasons, of which probably the most relevant are, on one side, an uneasiness of archaeologists in handing over –even partially– the interpretation process to machine-based judgment and, on the other, the fact that archaeological features can assume a near-unlimited assortment of shapes, sizes and spectral properties, which makes particularly challenging their auto-extraction. Thus, while (semi)automated and supervised procedures for feature extraction and processing are flourishing in a variety of fields, allowing for large swathes of landscapes to be simultaneously investigated, their application to archaeological and, more generally, cultural landscapes is still in its infancy.
A number of approaches in Feature extraction, Pattern Recognition, Pattern Matching, to name a few, now offer the opportunity to adopt (semi)automated feature detection and processing methods to identify potential archaeological features. These approaches can overcome the previous limitations of spectral and object-based methods and enable recognition of landscape patterns/features produced by a variety of diverse natural or artificial elements.
This session invites presentations showcasing computer-vision methods that are being used or developed to automatically identify landscape features and/or patterns on remote sensing imagery and it is –purposely– open to research employing broadly defined 'remote sensing data'. The session also welcomes controversial papers examining more broadly the subject from a theoretical point of view and addressing the topic from an antagonist angle.