Perception of Natural Scenes: Depth Estimation and the Influence of Colour

Abstract

Humans are remarkably efficient at understanding the visual structure of their environment. Within a fraction of a second, we can recognize a scene, identify the objects it contains, understand their spatial organization, and use this information to move and interact with the world with ease. This rapid and robust scene perception relies on the integration of multiple visual cues, including luminance, spatial structure, and colour. Colour is a pervasive feature of natural vision: it enriches our perceptual experience, facilitates object recognition, and provides information about illumination, surface properties, and materials. Although extensive research has examined how the visual system transforms retinal signals into perceptual experiences of colour, the functional role of colour in natural vision remains less well understood. In particular, the mechanisms by which colour contributes to scene understanding, and how it interacts with other sources of visual information such as luminance and spatial structure, are still underdeveloped. While many models of visual perception have historically prioritized luminance information over colour, largely because performance on numerous visual tasks remains relatively robust in the absence of colour, growing evidence suggests that colour contributes more fundamentally to the encoding, organization, and memory of natural scenes than previously assumed. The aim of this thesis is to examine the role of colour in the perception of natural scenes and to improve our understanding of how colour supports everyday visual perception. Chapter 1 provides an overview of colour vision research, reviewing key findings on how colour is perceived and processed in daily life. It also introduces major theories and research frameworks in natural scene perception, examining how colour is treated within these models and summarizing empirical evidence for the role of colour in natural scene understanding. Chapter 2 presents evidence that colour can act as a cue to depth perception in natural scenes, discussing mechanisms by which colour may support the perception of three-dimensional structure. Chapter 3 introduces a new scene segmentation dataset based on a large pool of annotators, in which images of natural scenes are segmented into objects and regions with detailed labels. The chapter also includes an experiment demonstrating how the dataset enables the study of influence of object types on depth judgments in natural scenes. Chapter 4 examines the role of colour in object and scene recognition under conditions of reduced spatial information, demonstrating that colour becomes particularly important at low spatial frequencies and when fine-grained luminance-based cues such as sharp edges are disrupted. Chapter 5 synthesizes the findings of the thesis, concluding that colour plays an important role in natural scene perception, particularly for depth perception and under degraded viewing conditions, and emphasizes the importance of studying colour vision in ecologically valid, natural settings.

More information

Identifiers

ORCID for Michaela Trescakova: orcid.org/0000-0002-2194-1243

ORCID for Erich Graf: orcid.org/0000-0002-3162-4233

ORCID for Christoph Witzel: orcid.org/0000-0001-9944-2420

ORCID for Wendy Adams: orcid.org/0000-0002-5832-1056

Catalogue record

Export record

Altmetrics