Texture Features and Image Texture Models

Abstract

Image texture is an important phenomenon in many applications of pattern recognition and computer vision. Hence, several models for deriving texture properties have been proposed and developed. Although there is no formal definition of image texture in the literature, image texture is usually considered the spatial arrangement of grayscale pixels in a neighborhood on the image. In this chapter, some widely used image texture methods for measuring and extracting texture features will be introduced. These textural features can then be used for image texture classification and segmentation. Specifically, the following methods will be described: (1) the gray-level co-occurrence matrices (GLCM) which is one of the earliest methods for image texture extraction, (2) Gabor filters, (3) wavelet transform (WT) model and its extension, (4) autocorrelationfunction, (5) Markov random fields (MRF), (6) fractal features, (7) variogram, (8) local binary pattern (LBP), and (9) texture spectrum (TS). LBP has been frequently used for image texture measure. MRF is a statistical model which has been well studied in image texture analysis and other applications. There is one common property associated with these methods and models which use the spatial relationship for texture measurement and classification.