Computing Science, School of

Receive updates for this collection

Experiential Reasoning

Author: 
Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
1992-03
Abstract: 

At the time I developed Whisper in 1976 1 encountered extremely skeptical---often downright hostile---audiences and as a result I (mistakenly) published only an expurgated account of my thoughts about analog representations. 1 I am delighted that times have changed a bit and this symposium seems like an appropriate forum for a little speculation and re-evaluation.

Document type: 
Conference presentation
File(s): 

Learning Color Constancy

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
1996-11
Abstract: 

We decided to test a surprisingly simple hypothesis; namely, that the relationship between an image of a scene and the chromaticity of scene illumination could be learned by a neural network. The thoughts was that if this relationship could be extracted by a neural network, then the trained network would be able to determine a scene's Illuminant from its image, which would then allow correction of the image colors to those relative to a standard illuminance, thereby providing color constancy. Using a database of surface reflectances and illuminants, along with the spectral sensitivity functions of our camera, we generated thousands of images of randomly selected illuminants lighting 'scenes' of 1 to 60 randomly selected reflectances. During the learning phase the network is provided the image data along with the chromaticity of its illuminant. After training, the network outputs (very quickly) the chromaticity of the illumination given only the image data. We obtained surprisingly good estimates of the ambient illumination lighting from the network even when applied to scenes in our lab that were completely unrelated to the training data.

Document type: 
Conference presentation
File(s): 

Bootstrapping Color Constancy

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
1999-01
Abstract: 

Bootstrapping provides a novel approach to training a neural network to estimate the chromaticity of the illuminant in a scene given image data alone. For initial training, the network requires feedback about the accuracy of the network’s current results. In the case of a network for color constancy, this feedback is the chromaticity of the incident scene illumination. In the past1, perfect feedback has been used, but in the bootstrapping method feedback with a considerable degree of random error can be used to train the network instead. In particular, the grayworld algorithm2, which only provides modest color constancy performance, is used to train a neural network which in the end performs better than the grayworld algorithm used to train it.

Document type: 
Conference presentation
File(s): 

Tuning Retinex Parameters

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2004-01
Abstract: 

Our goal is to understand how the Retinex parameters affect the predictions of the model. A simplified Retinex computation is specified in the recent MATLAB™ implementation; however, there remain several free parameters that introduce significant variability into the model’s predictions. We extend previous work on specifying these parameters. In particular, instead of looking for fixed values for the parameters, we establish methods that automatically determine values for them based on the input image. These methods are tested on the McCann-McKee-Taylor asymmetric matching data, along with some previously unpublished data that include simultaneous contrast targets.

Document type: 
Article
File(s): 

A Large Image Database for Color Constancy Research

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2003-11
Abstract: 

We present a study on various statistics relevant to research on color constancy. Many of these analyses could not have been done before simply because a large database for color constancy was not available. Our image database consists of approximately 11,000 images in which the RGB color of the ambient illuminant in each scene is measured. To build such a large database we used a novel set-up consisting of a digital video camera with a neutral gray sphere attached to the camera so that the sphere always appears in the field of view. Using a gray sphere instead of the standard gray card facilitates measurement of the variation in illumination as a function of incident angle. The study focuses on the analysis of the distribution of various illuminants in the natural scenes and the correlation between the rg-chromaticity of colors recorded by the camera and the rg-chromaticity of the ambient illuminant. We also investigate the possibility of improving the performance of the naïve Gray World algorithm by considering a sequence of consecutive frames instead of a single image. The set of images is publicly available and can also be used as a database for testing color constancy algorithms.

Document type: 
Conference presentation
File(s): 

Estimating Illumination Chromaticity via Support Vector Regression

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2004-11
Abstract: 

The technique of support vector regression is applied to the problem of estimating the chromaticity of the light illuminating a scene from a color histogram of an image of the scene. Illumination estimation is fundamental to white balancing digital color images and to understanding human color constancy. Under controlled experimental conditions, the support vector method is shown to perform better than the neural network and color by correlation methods.

Document type: 
Conference presentation
File(s): 

Optimal Linear RGB-to-XYZ Mapping for Color Display Calibration

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2004-11
Abstract: 

Color display calibration, in part, involves mapping input RGB values to corresponding output values in a standardized color space such as CIE XYZ. A linear model for RGB-to- XYZ mapping is based on a 3-by-3 linear transformation matrix T mapping data from (linearized) RGB to XYZ. Such a mapping is often determined by least squares regression on the difference between predicted and measured XYZ values. However, since displays are calibrated for viewing by human observers, it likely would be better to optimize relative to a perceptually uniform color space. Two new methods are proposed which optimize the total error relative to CIELAB or CIEDE2000. The first method uses weighted least squares with weights based on the rate of change of CIELAB coordinates as a function of change in XYZ. The second method uses Nedler-Mead nonlinear optimization to minimize directly in CIELAB or CIEDE200. Experiments based on calibrating 2 CRT monitors, 3 LCD monitors and 2 LCD projectors show significantly better results than the standard least squares calibration.

Document type: 
Conference presentation
File(s): 

Retinex in MATLAB™

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2004-01
Abstract: 

Many different descriptions of Retinex methods of lightness computation exist. We provide concise MATLAB™ implementations of two of the spatial techniques of making pixel comparisons. The code is presented, along with test results on several images and a discussion of the results. We also discuss the calibration of input images and the postRetinex processing required to display the output images.

Document type: 
Article
File(s): 

Nonlinear RGB-to-XYZ Mapping for Device Calibration

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2005-11
Abstract: 

We introduce a new non-linear method for RGB-to-XYZ color calibration based on the technique of thin plate splines. Originally, thin plate splines were designed for deformable matching between 2-dimensional images for object recognition. We use 3-dimensional thin plate splines to map between sets of RGB device coordinates and corresponding sets of CIE XYZ coordinates. Tests calibrating several displays as well as a camera show thin plate spline calibration to be more accurate than existing linear or non-linear calibration methods.

Document type: 
Conference presentation
File(s): 

Illumination Estimation from Dichromatic Planes

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2005-11
Abstract: 

Adopting the dichromatic reflection model under the assumption of neutral interface reflection, the color of the illuminating light can be estimated by intersecting the planes that the color response of two or more different materials describe. From the color response of any given region, most approaches estimate a single plane on the assumption that only a single material is imaged. This assumption, however, is often violated in cluttered scenes. In this paper, rather than a single planar model, several coexisting planes are used to explain the observed color response. In estimating the illuminant, a set of candidate lights is assessed for goodness of fit given the assumed number of coexisting planes. The candidate light giving the minimum error fit is then chosen as representative of the scene illuminant. The performance of the proposed approach is explored on real images.

Document type: 
Conference presentation
File(s):