Using Image Depth Variables as Predictors of Visual Quality

Abstract

Depth is an important part of our understanding and appreciation of landscape. Attempts made to measure attributes of photographs of landscapes and to develop predictive models of scenic beauty have either failed to include depth or included it by subjective separation of the foreground, mid-ground, and background within the photograph. The process of measurement from photographs is also limiting because it does not provide a mechanism for readily assessing the effect on visual qualities of proposed landscape changes. The process described here requires a detailed digital model of the environment but is then able to generate variables relating to land-cover type, their depth, and patterns of depth in any view direction. Among these variables are several which are closely correlated with scenic beauty and which can together make a good predictive model. Computer systems, which include hardware implementations of three-dimensional visible surface algorithms, offer the possibility of very fast generation of scenic beauty estimates from any location. Such estimates are among the requirements for autonomous agent-based modelling of landscape behaviour.