Ph.D. thesis: M.J.
Chantler
"The effect of variation in illuminant direction on texture classification",
Dept. Computing and Electrical Engineering, Heriot-Watt University,
August 1994.
Entire Thesis PDF
Abstract
Texture analysis has been an extremely active and fruitful area of research
over the past twenty years. Many advances have been made, but the effect
of variation in lighting conditions on automated texture classification
and segmentation has received little attention. This thesis shows that
the direction of the illuminant is an important factor that should be taken
into account when analyzing images of three-dimensional texture.
A frequency domain model is presented which predicts that both the directional
characteristics and the variance of images of three-dimensional texture
can be affected by changes in illuminant vector. Results of simulations
and laboratory experiments support these predictions.
The responses of three sets of texture measures are analyzed using a
test set of isotropic and directional textures. The results show that the
feature measures' outputs are affected by changes in illuminant direction.
These changes are also shown to significantly increase the error rates
of statistical classifiers implemented using the three feature sets. Normalization
of images is shown to reduce the error rates in some cases.
The frequency domain model of image texture is further developed using
empirical data and the resulting model used to design a set of tilt-compensation
filters. These filters are used to pre-process images to reduce the effects
of changes in the angle of tilt of the illuminant. Application of the filters
to the test image set reduced the classification errors associated with
directional textures.
Table of Contents
Preliminary Pages (chapter0.pdf)
Chapter 1 Introduction (chapter1.pdf)
1.1. Motivation
1.2. Scope of the research
1.3. Thesis organization
1.4. Original work
Chapter 2 Image models of topological texture (chapter2.pdf)
2.1. Review
2.1.1. Texture synthesis
2.1.2. Texture analysis - segmentation and classification
2.1.3. Texture analysis - shape from texture
2.1.4. Scattering theory
2.1.5. Summary
2.2. An image model of topological texture
2.2.1. A fractal based image model
2.2.2. Implications for texture analysis
2.3. Conclusions
Chapter 3 An investigation into an image model of topological texture
(chapter3.pdf)
3.1. The response of image texture to changes in surface relief
3.1.1. Image generation
3.1.2. The power roll-off factor
3.1.3. Large slope angles
3.1.4. Shadowing
3.1.5. Summary of surface response results
3.2. The tilt angle response of image texture
3.2.1. Low slope angles
3.2.2. Large slope angles
3.2.3. Shadowing
3.2.4. Four physical textures
3.2.5. Summary of tilt response investigation
3.3. The slant angle response of image texture
3.3.1. Low slope angles
3.3.2. Large slope angles and shadowing
3.3.3. Experimental results : slant response
3.3.4. Radial shape - slant angle response
3.3.5. Summary of slant response investigation
3.4. Conclusions
3.4.1. Implications for texture classification
Chapter 4 Texture features: review and selection (chapter4.pdf)
4.1. Definition of segmentation, classification and feature measure.
4.2. Surveys
4.3. Model-based features
4.3.1. Fractal models
4.3.2. Autoregressive models
4.3.3. Fractional differencing models
4.3.4. Markov random fields
4.4. Non-model-based features
4.4.1. Grey-level co-occurrence and other related features
4.4.2. Laws' texture energy filters
4.4.3. Frequency domain methods
4.4.4. Gabor filters
4.5. Comparative studies
4.5.1. A league table of feature measures
4.6. Rotation invariance
4.6.1. Omnidirectional feature measures
4.6.2. Rotation invariant directional feature measures
4.7. Conclusion
Chapter 5 Texture features and illumination (chapter5.pdf)
5.1. Laws' masks
5.1.1. Frequency response
5.1.2. Tilt angle response
5.1.3. Slant angle response
5.1.4. Summary
5.2. Co-occurrence matrices
5.2.1. An alternative formulation
5.2.2. Frequency response
5.2.3. Tilt angle response
5.2.4. Slant angle response
5.2.5. Summary
5.3. Linnett's operator
5.3.1. Frequency response
5.3.2. Tilt angle response
5.3.3. Slant angle response
5.3.4. Summary - Linnett's operator
5.4. Metrics for class separation and sensitivity to illuminant variation
5.5. Conclusions
Chapter 6 Classification (chapter6.pdf)
6.1. Supervised statistical classification
6.1.1. Discriminant theory
6.1.2. Supervised classification of test textures
6.2. The effect of illuminant variation on classification
6.2.1. Discrimination between illumination conditions
6.2.2. Slant response
6.2.3. Tilt response
6.2.4. Summary of illuminant variation investigation
6.3. Compensation for illuminant tilt variation
6.4. Frequency domain tilt-compensation
6.4.1. An improved frequency domain model
6.4.2. Filter implementation
6.4.3. Effect of tilt-compensation on features
6.4.4. Effect of tilt-compensation on classification
6.5. Conclusions
Chapter 7 Summary and conclusions (chapter7.pdf)
7.1. Summary
7.2. Conclusions
Appendix A (appendix-A.pdf)
References (references.pdf)