PaperlayerView exact revision on arXiv

Source-linked AI summary

Texture image analysis and texture classification methods - A review

Laleh Armi, Shervan Fekri-Ershad

arXiv:1904.06554v1cs.CV

TL;DR

Texture analysis seeks useful descriptions of spatial intensity variation for computer-vision tasks, but methods must handle challenges such as noise, rotation, and computational complexity. This review organizes established approaches into four categories, examines combinational methods, and surveys classifiers and benchmark datasets. It emphasizes discrimination performance, complexity, and resistance to imaging challenges across the reviewed descriptors.

  • Problem

    Texture classification must describe spatial intensity patterns despite rotation, noise, scale, viewpoint, and illumination differences that can reduce accuracy.

  • Method

    The paper reviews statistical, structural, model-based, transform-based, and combinational texture-analysis methods, plus classifiers and benchmark datasets.

  • Results

    The review identifies diverse texture descriptors and reports that combinational methods span multiple categories, with statistical and transform-based methods especially prevalent.

  • Takeaways & Limitations

    Texture-analysis methods should be considered in relation to discrimination performance, computational complexity, and resistance to noise and rotation.

Abstract

from arXiv · show

Tactile texture refers to the tangible feel of a surface and visual texture refers to see the shape or contents of the image. In the image processing, the texture can be defined as a function of spatial variation of the brightness intensity of the pixels. Texture is the main term used to define objects or concepts of a given image. Texture analysis plays an important role in computer vision cases such as object recognition, surface defect detection, pattern recognition, medical image analysis, etc. Since now many approaches have been proposed to describe texture images accurately. Texture analysis methods usually are classified into four categories: statistical methods, structural, model-based and transform-based methods. This paper discusses the various methods used for texture or analysis in details. New researches shows the power of combinational methods for texture analysis, which can't be in specific category. This paper provides a review on well known combinational methods in a specific section with details. This paper counts advantages and disadvantages of well-known texture image descriptors in the result part. Main focus in all of the survived methods is on discrimination performance, computational complexity and resistance to challenges such as noise, rotation, etc. A brief review is also made on the common classifiers used for texture image classification. Also, a survey on texture image benchmark datasets is included.

1 Faculty of Computer Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Iran

This passage lists keywords covering texture imagery, analysis, classification, feature extraction, image processing, Local Binary Patterns, and benchmark datasets.

  • The paper concerns texture images, texture analysis, and texture classification.
  • Feature extraction and image processing are central methodological themes.
  • Local Binary Patterns and benchmark texture image datasets are specifically highlighted.

I. INTRODUCTION

The introduction defines texture as spatial brightness variation and frames texture analysis as a broad computer-vision problem. It presents a review organized around method categories, combined approaches, classifiers, datasets, and practical challenges including noise, rotation, complexity, and discrimination.

  • Texture describes spatial variation in pixel brightness and repeated intensity patterns across images.The paper distinguishes tactile texture from visual texture and uses repetitive surfaces as examples.
  • Texture classification assigns an unknown sample image to a predefined texture class.
  • Classification process: Classification commonly uses feature extraction followed by comparison of test-image features with training imagery and a classifier.
  • Review scope: The review covers statistical, structural, model-based, and transform-based methods, alongside combined methods, classifiers, datasets, and method trade-offs.It discusses advantages and disadvantages while emphasizing discrimination performance, computational complexity, and robustness challenges.
  • Statistical methods: Statistical methods derive feature vectors from pixel-intensity distributions, with higher-order statistics incorporating dependencies among multiple pixels.The co-occurrence matrix is presented as a second-level histogram.
  • Statistical methods: Haralick co-occurrence matrices count brightness-pair occurrences at a displacement and direction, enabling statistical texture properties such as contrast and homogeneity.For an L-level image, the matrix is L×L.
  • Model-based methods: Model-based methods represent textures through models such as autoregressive, Markov, Gibbs, hidden Markov, and fractal models.Fractal models are described as capturing coarseness, harshness, and self-similarity.

III. Combinational state-of-the-art Texture Analysis Algorithms

The reviewed combinational texture-analysis methods combine multiple descriptors, transforms, statistical features, or classifiers to improve robustness and discrimination. Examples include GLCM extensions, SEGL, BGP, energy-based features, and SVD-DWT with optimized SVM classification.

  • Combinational methods: The reviewed GLCM extensions include pyramid representations and multi-directional matrices, while other approaches combine DWT and SVD-DWT features.The pyramid approach represents an image in five layers and computes co-occurrence matrices in four directions.
  • Combinational methods: SEGL combines LBP, GLCM, edge detection, and seven statistical features, producing 56 attributes from eight directional co-occurrence matrices.The features are entropy, energy, contrast, homogeneity, correlation, average, and variance.
  • Combinational methods: Energy-based texture classification combines three co-occurrence operators, LBP, and edge features, then forms a vector from differences between original and processed energies.Original energy and processing energy are explicitly used in the feature-vector construction.
  • Combinational methods: BGP improves LBP by applying eight directionally different Gabor filters before generating a binary pattern.The resulting eight-dimensional image is processed similarly to the spatially insensitive LBP.
  • Robustness: BGP is reported as resistant to rotation because rotational shifts are selected by their maximum value before histogram construction and nearest-neighbor classification.The method produces the same result for images shown at different rotation angles.
  • Combinational methods: The SRITCSD method enhances textures with SVD, extracts one-level four-subband DWT features, and uses PSO to optimize SVM features and parameters.The four sub-bands are LL1, LH1, HL1, and HH1, producing eight feature values.

IV. Benchmark Texture Image Datasets

The survey describes benchmark datasets spanning scanned grayscale textures, natural and artificial textures, color scenes, reflectance variations, and high-resolution material images. Their class counts, image conditions, and viewpoints vary substantially, supporting comparisons under different texture-analysis settings.

  • Dataset role: Standard texture datasets provide common evaluation material and enable comparisons between texture-analysis algorithms.The survey emphasizes their role in assessing proposed systems and algorithms.
  • Grayscale and natural textures: Brodatz contains 112 640×640 grayscale texture images but is limited to a single illumination and viewing direction per texture.The images use 8-bit intensity with 256 gray values.
  • Color and scene textures: VisTex contains 640 128×128 color images across 40 classes, including homogeneous textures and real-world scenes with multiple textures.The dataset was designed to represent real-world conditions.
  • Reflectance and viewpoint variation: CUReT contains 61 classes with 92 images each at 200×200 resolution, captured under varying illumination and viewpoints.The survey describes it as more suitable than Brodatz for evaluating proposed algorithms.
  • Material datasets: KTH-TIPS2 has 11 classes, four materials per class, 432 images per class, and 4,752 images overall.Different colors can aid discrimination for some materials, while different classes can share similar colors.
  • Material datasets: The FMD dataset contains 10 real-world material categories with 100 images per category, split into 50 close-ups and 50 regular views.Categories include fabric, foliage, glass, leather, metal, paper, plastic, stone, water, and wood.

V. Classification Phase in Texture Classification Methods:

The classification phase trains supervised models on extracted texture features, then assigns test images to classes using algorithms such as KNN, SVM, Naive Bayes, decision trees, and neural networks. The review compares texture methods across accuracy, complexity, and robustness, highlighting combined approaches and persistent noise sensitivity.

  • Classification workflow: Texture classification uses supervised learning to build a model from feature vectors paired with class information.The trained model compares test-image features with training imagery to determine the class.
  • K-Nearest Neighbor: KNN assigns a test image the majority label among its k nearest training samples in feature space.Larger k can reduce noise effects but may increase errors for classes with few samples.
  • Support Vector Machine: SVM separates complex patterns using kernels and extends binary classification to multiple classes through one-versus-one or one-versus-rest strategies.The review illustrates both the binary-separation basis and multiclass extensions of SVM.
  • Other classifiers: Naive Bayes, decision trees, and neural networks provide alternative supervised classifiers for texture-image categorization.Decision trees infer classes through root-to-leaf rules, while multilayer perceptrons update weights through back-propagation.
  • Comparison of methods: Reported results include 99.04% for KNN on Brodatz, 99.83% for KNN on VisTex, and 99.42% for ELM on Brodatz.These values are presented in the review’s comparison of state-of-the-art texture-analysis methods.
  • Comparison of methods: The review reports that most methods address rotation, while many remain sensitive to noise; combined statistical and transform-based methods are common.Reported examples include rotation-invariant descriptors and methods combining spatial and frequency-domain features, but noise sensitivity remains frequent.

Authors Biography

The authors are researchers in artificial intelligence and computer engineering whose interests include image processing, computer vision, texture analysis, and texture classification.

  • Laleh Armi: Laleh Armi’s research interests include image processing, computer vision applications, texture analysis, texture classification, and image descriptors.She received an M.Sc. in computer engineering with an artificial-intelligence major in 2017.
  • Shervan Fekri-Ershad: Shervan Fekri-Ershad’s research interests include image-processing applications and visual information.He holds a Ph.D. in artificial intelligence and joined the faculty of computer engineering at Najafabad branch.
Loading 1904.06554v1…