Texture Analysis

Texture Analysis

Texture analysis is an important topic in image processing and has been used in many applications including automated inspection, image retrieval, remote sensing, object recognition, image matching, and medical image analysis. Image textures are defined as visual patterns appearing in images. Textures are all around us. We can see on on walls, carpets, fingerprints, and medical images such as Ultrasound images and MRI images.

Texture analysis methods use chromatic and structural characteristics of images to characterize textures. There are many different methods for texture analysis; however, they can be categorized into four general groups.

  1. Statistical methods: The main motivation behind these methods is based on the fact that the human visual system uses statistical features to distinguish textures. Some well-known methods are the gray level Aura matrices (GLCM), the gray level Aura matrices (GLAM), and Local Binary Patterns (LBP).

  2. Structural-based methods: These methods decompose textures into elements known as primitives or texels. The primitives and their spatial arrangements are used to characterize textures. In general, the structural-based methods are better suited for textures with large structures (macrostructure) and do not work well on stochastic textures and microtextures.

  3. Probability models: Some well-known models are Markov Random Field (MRF), Auto Regressive (AR) model , and Gibbs random field. The key issue in these models is how to choose the correct model for a given texture and how to ectively map a texture into the selected probability model.

  4. Filter-based methods: These methods apply filters on images in either spatial or frequency domain. Windowed Fourier filters, wavelet transform, Gabor filter banks, and spatial filters are used by different methods in this category.

Extracting rotation invariant textural features is an interesting and popular topic in computer vision and image processing. Rotation invariant features can be used directly or indirectly in different applications. Direct applications include those that need only rotation invariance (e.g., rotation invariant texture classification). In advanced applications such as sparse texture classification or image matching the rotation invariant features are employed to obtain affine invariance. In these applications, a set of elliptical regions are found using region detection methods such as Laplacian, Hessian, or Harris region detectors, each elliptical region is normalized to a circle, and rotation invariant features are computed on the normalized circle.

The most famous rotaiont invariant texture method is LBP. The following are the sagittal and axial T1 brain images and the corresponding LBP patterns. The 10 different LBPriu patterns are shown in 10 different colors.