1 Introduction.- 1.1 Scope of the book.- 1.1.1 Defining texture.- 1.1.2 Problem definition.- 1.2 Importance of texture.- 1.3 Potential applications of this research.- 1.4 Issues in automated process control involving computer vision.- 1.4.1 Background.- 1.4.2 Problems involving texture.- 1.4.3 The desired solution.- 1.5 A taxonomy for texture.- 1.6 Outline.- 2 Computing oriented texture fields.- 2.1 Introduction.- 2.2 Background.- 2.3 Oriented Texture Fields.- 2.4 Experimental Methods.- 2.4.1 A best estimate for dominant local orientation.- 2.4.2 Derivation using the moment method.- 2.4.3 Squaring the Gradient Vectors: Kass and Witkin's scheme.- 2.4.4 Inverse Arctangent.- 2.4.5 Flow Orientation Coherence.- 2.4.6 The effect of varying ?1 on the estimate of dominant orientation.- 2.5 Experimental Results.- 2.5.1 Comparing Calculations for Orientation.- 2.5.2 Comparing Measures of Coherence.- 2.6 Analyzing texture at different scales.- 2.7 Processing of the intrinsic images.- 2.7.1 The coherence intrinsic image.- 2.7.2 The angle intrinsic image.- 2.8 Conclusions.- 3 The analysis of oriented textures through phase portraits.- 3.1 Introduction.- 3.1.1 Overview of approach.- 3.2 Background.- 3.2.1 Related Research.- 3.2.2 Potential Applications.- 3.2.3 Obtaining the intrinsic images.- 3.2.4 Problems in post-processing.- 3.3 Geometric theory of differential equations.- 3.3.1 Preliminary definitions.- 188.8.131.52 One dimensional case.- 184.108.40.206 Two dimensional case.- 3.3.2 Linear systems.- 220.127.116.11 Case 1: Matrix A is non-singular.- 18.104.22.168 Case 2: Matrix A is singular.- 22.214.171.124 Affine transformations.- 3.3.3 Relevance of the theory.- 3.3.4 The perceptual significance of flow-like patterns.- 3.4 Experimental Methods.- 3.4.1 A Distance Measure.- 3.4.2 Non-linear least squares fitting.- 3.4.3 Implementation details.- 126.96.36.199 Generating the orientation field O2.- 188.8.131.52 Normalization of orientation fields.- 184.108.40.206 Uniqueness of the solution.- 220.127.116.11 Real Images.- 3.4.4 Segmentation.- 18.104.22.168 A measure for closeness of fit.- 22.214.171.124 Selecting the size of the window.- 3.4.5 Locating fixed points.- 3.4.6 Reconstructing the original texture.- 3.5 Experimental Results.- 3.5.1 Flow past an inclined plate.- 3.5.2 Image of secondary streaming.- 3.5.3 Image of wood knots.- 3.5.4 Image of complex flow.- 3.5.5 Image of vortex flow.- 3.5.6 Analysis of a resist gel defect.- 3.5.7 Analysis of textured paint brush strokes.- 3.6 Experiments with noise addition.- 3.7 A related model from fluid flow analysis.- 3.7.1 A comparison between the fluid motion viewpoint and the phase portrait viewpoint.- 3.7.2 Classification of velocity fields.- 3.7.3 Importance of divergence and curl.- 3.8 Discussion.- 3.8.1 Extensions to three dimensions.- 3.9 Conclusion.- 4 Analyzing strongly ordered textures.- 4.1 Introduction.- 4.2 Extraction of primitives.- 4.2.1 Edge based features.- 4.2.2 Region based features.- 126.96.36.199 Thresholding the response to ?2G.- 188.8.131.52 Solving for the response to a disk at multiple scales.- 4.3 Extracting structure from primitives.- 4.3.1 Syntactic approaches.- 4.3.2 Nearest Neighbor Histograms.- 4.3.3 Using co-occurrence matrices.- 4.3.4 Graph based approaches.- 4.4 Models for strongly ordered textures.- 4.4.1 Directions for further research.- 4.5 Symbolic descriptions: models from petrography.- 4.5.1 Description of primitives.- 184.108.40.206 Terminology from petrology.- 4.5.2 Description of placement of primitives.- 4.6 Frieze groups and wallpaper groups.- 4.6.1 Background.- 4.6.2 Preliminary definitions.- 4.6.3 Frieze groups.- 4.6.4 Wallpaper groups.- 4.7 Implications for computer vision.- 4.8 Summary.- 5 Disordered textures.- 5.1 Statistical measures for disordered textures.- 5.1.1 Computing the entropy as a measure for disorder.- 5.2 Describing disordered textures by means of the fractal dimension.- 5.2.1 Advantages.- 5.3 Computing the fractal dimension.- 5.3.1 Using the expected values of intensity differences.- 5.3.2 The reticular cell counti
A central issue in computer vision is the problem of signal to symbol transformation. In the case of texture, which is an important visual cue, this problem has hitherto received very little attention. This book presents a solution to the signal to symbol transformation problem for texture. The symbolic de- scription scheme consists of a novel taxonomy for textures, and is based on appropriate mathematical models for different kinds of texture. The taxonomy classifies textures into the broad classes of disordered, strongly ordered, weakly ordered and compositional. Disordered textures are described by statistical mea- sures, strongly ordered textures by the placement of primitives, and weakly ordered textures by an orientation field. Compositional textures are created from these three classes of texture by using certain rules of composition. The unifying theme of this book is to provide standardized symbolic descriptions that serve as a descriptive vocabulary for textures. The algorithms developed in the book have been applied to a wide variety of textured images arising in semiconductor wafer inspection, flow visualization and lumber processing. The taxonomy for texture can serve as a scheme for the identification and description of surface flaws and defects occurring in a wide range of practical applications.
This book solves the signal to symbol transformation problem for texture, an important issue in computer vision. The algorithms developed here have been applied to a wide variety of textured images.