Cue combination of colour and luminance in edge detection

Sharman, Rebecca J. (2014) Cue combination of colour and luminance in edge detection. PhD thesis, University of Nottingham.

PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (2MB) | Preview


Much is known about visual processing of chromatic and luminance information. However, less is known about how these two signals are combined. This thesis has three aims to investigate how colour and luminance are combined in edge detection. 1) To determine whether presenting colour and luminance information together improves performance in tasks such as edge localisation and blur detection. 2) To investigate how the visual system resolves conflicts between colour and luminance edge information. 3) To explore whether colour and luminance edge information is always combined in the same way.

It is well known that the perception of chromatic blur can be constrained by sharp luminance information in natural scenes. The first set of experiments (Chapter 3) quantifies this effect and demonstrates that it cannot be explained by poorer acuity in processing chromatic information, higher contrast of luminance information or differences in the statistical structure of colour and luminance information in natural scenes. It is therefore proposed that there is a neural mechanism that actively promotes luminance information.

Chapter 4 and Experiments 5.1 and 5.3 aimed to investigate whether the presence of both chromatic and luminance information improves edge localisation performance. Participant performance in a Vernier acuity (alignment) task was compared to predictions from three models; ‘winner takes all’, unweighted averaging and maximum likelihood estimation (a form of weighted averaging). Despite several attempts to differentiate the models we failed to increase the differences in model predictions sufficiently and it was not possible to determine whether edge localisation was enhanced by the presence of both cues.

In Experiment 5.4 we investigated how edges are localised when colour and luminance cues conflict, using the method of adjustment. Maximum likelihood estimation was used to make predictions based on measurements of each cue in isolation. These predictions were then compared to observed data. It was found that, whilst maximum likelihood estimation captured the pattern of the data, it consistently over-estimated the weight of the luminance component. It is suggested that chromatic information may be weighted more heavily than predicted as it is more useful for detecting object boundaries in natural scenes.

In Chapter 6 a novel approach, perturbation discrimination, was used to investigate how the spatial arrangement of chromatic and luminance cues, and the type of chromatic and luminance information, can affect cue combination. Perturbation discrimination requires participants to select the grating stimulus that contains spatial perturbation. If one cue dominated over the other it was expected that this would be reflected by masking and increased perturbation detection thresholds. We compared perturbation thresholds for chromatic and luminance defined line and square-wave gratings in isolation and when presented with a mask of the other channel and other grating type. For example, the perturbation threshold for a luminance line target alone was compared to the threshold for a luminance line target presented with a chromatic square-wave target. The introduction of line masks caused masking for both combinations. Introduction of an achromatic square-wave mask had no effect on perturbation thresholds for chromatic line targets. However, the introduction of a chromatic square-wave mask to luminance line targets improved perturbation discrimination performance. This suggests that the perceived location of the chromatic edges is determined by the location of the luminance lines.

Finally, in Chapter 7, we investigated whether chromatic blur is constrained by luminance information in bipartite edges. Earlier in the thesis we demonstrated that luminance information constrains chromatic blur in natural scenes, but also that chromatic information has more influence than expected when colour and luminance edges conflict. This difference may be due to differences in the stimuli or due to differences in the task. The luminance masking effect found using natural scenes was replicated using bipartite edges. Therefore, the finding that luminance constrains chromatic blur is not limited to natural scene stimuli. This suggests that colour and luminance are combined differently for blur discrimination tasks and edge localisation tasks.

Overall we can see that luminance often dominates in edge perception tasks. For blur discrimination this seems to be because the mechanisms differ. For edge localisation it might be simply that luminance cues are often higher contrast and, when this is equated, chromatic cues are actually a good indicator of edge location.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Peirce, J.
McGraw, P.
Keywords: visual processing, chromatic information, luminance information, edge detection, chromatic blur
Subjects: B Philosophy. Psychology. Religion > BF Psychology
Faculties/Schools: UK Campuses > Faculty of Science > School of Psychology
Item ID: 14029
Depositing User: EP, Services
Date Deposited: 22 Dec 2014 13:51
Last Modified: 17 Oct 2017 04:50

Actions (Archive Staff Only)

Edit View Edit View