Fractal dimensions of landscape images as predictors of landscape preference

Abstract

Many studies of natural landscape preference have demonstrated that qualities such as 'complexity' and 'naturalness' are associated with preference, but have struggled to define the key characteristics of these qualities. Recently, the development of software programs and digital techniques has offered researchers new ways of quantifying the landscape qualities associated with preference. Among them fractal geometry offers the most promising approach. Fractals have been defined as mathematical models of organic objects and patterns as opposed to the straight lines and perfect circles of Euclidean geometry found in man-made environments. Fractal patterns are mainly characterized by their dimension, which could be described as a statistical quantification of complexity. By applying this mathematical concept to digital images, several studies claim to have found a correlation between the fractal dimensions of a set of images and the images' preference ratings. Such studies have particularly focussed on demonstrating support for the hypothesis that patterns with a fractal dimension of around 1.3 induce better responses than others. However, much of this research so far has been carried out on abstract or computer-generated images. Furthermore, the most commonly used method of fractal analysis, the box-counting method, has many limitations in its application to digital images which are rarely addressed. The aim of this thesis is to explore empirically the suggestion that landscape preference could be influenced by the fractal characteristics of landscape photographs. The first part of this study was dedicated to establishing the robustness and validity of the box-counting method, and apply it to landscape images. One of the main limitations of the box-counting method is its need for image pre-processing as it can only be applied to binary (black and white) images. Therefore, to develop a more reliable method for fractal analysis of landscapes, it was necessary to compare different methods of image segmentation, i.e the reduction of greyscale photographs into binary images. Each method extracted a different structure from the original photograph: the silhouette outline, the extracted edges, and three different thresholds of greyscale. The results revealed that each structure characterized a different aspect of the landscape: the fractal dimension of the silhouette outline could quantify the height of the vegetation, while the fractal dimension of the extracted edges characterized complexity. The second part of the study focused on collecting preference ratings for the landscape images previously analysed, using an online survey disseminated in France and the UK. It was found that different groups of participants reacted differently to the fractal dimensions, and that some of those groups were significantly influenced by those characteristics while others were not. Unexpectedly, the variable most correlated with preference was the fractal dimension of the image's extracted edges, although this variable's predictive power was relatively low. The study concludes by summarising the issues involved in estimating the fractal dimensions of landscapes in relation to human response. The research offers a set of reliable and tested methods for extracting fractal dimensions for any given image. Using such methods, it produces results which challenge previous hypotheses and findings in relation to fractal dimensions that predict human preference, identifying gaps in understanding and promising future areas of research.