Textural Features for Image Classification¶
Why this mattered¶
Haralick, Shanmugam, and Dinstein made texture a computable object for image classification. Before this paper, texture was widely recognized as visually important, but it was difficult to express in a compact, reproducible form that could be fed into statistical classifiers. The paper’s gray-level co-occurrence matrices shifted attention from individual pixel values to spatial relationships between gray tones: how often particular intensity pairs occur at defined offsets and directions. That move gave researchers a practical vocabulary for measuring image structure through features such as contrast, entropy, homogeneity, and correlation, making texture analysis portable across microscopy, aerial photography, and satellite imagery.
The significance was not that the reported accuracies were definitive, but that the same feature construction worked across very different visual domains. The paper showed that local spatial dependence could support supervised recognition of rock micrographs, land-use regions in aerial photos, and multispectral satellite scenes. This helped establish a template for pre-deep-learning computer vision: design interpretable features, estimate them from image regions, and classify them with relatively simple decision rules. In doing so, it made automated texture-based remote sensing, biomedical image analysis, materials inspection, and industrial vision more systematic and comparable.
Its influence persisted because the Haralick feature family became a standard baseline and building block for decades of image analysis. Later methods such as filter banks, local binary patterns, wavelets, and learned convolutional features pursued richer or more adaptive descriptions of spatial pattern, but they addressed the same core problem the paper formalized: representing visual texture in a way that supports recognition. Even after deep learning displaced much handcrafted feature engineering, gray-level co-occurrence features remained important as interpretable descriptors, especially in domains such as radiomics and remote sensing where small datasets, explainability, and physically meaningful measurements still matter.
Abstract¶
Texture is one of the important characteristics used in identifying objects or regions of interest in an image, whether the image be a photomicrograph, an aerial photograph, or a satellite image. This paper describes some easily computable textural features based on gray-tone spatial dependancies, and illustrates their application in category-identification tasks of three different kinds of image data: photomicrographs of five kinds of sandstones, 1:20 000 panchromatic aerial photographs of eight land-use categories, and Earth Resources Technology Satellite (ERTS) multispecial imagery containing seven land-use categories. We use two kinds of decision rules: one for which the decision regions are convex polyhedra (a piecewise linear decision rule), and one for which the decision regions are rectangular parallelpipeds (a min-max decision rule). In each experiment the data set was divided into two parts, a training set and a test set. Test set identification accuracy is 89 percent for the photomicrographs, 82 percent for the aerial photographic imagery, and 83 percent for the satellite imagery. These results indicate that the easily computable textural features probably have a general applicability for a wide variety of image-classification applications.
Related¶
- enables → Radiomics: Images Are More than Pictures, They Are Data — Haralick texture features introduced quantitative image descriptors that radiomics generalized into high-throughput medical-image feature extraction.
- cite ← Radiomics: Images Are More than Pictures, They Are Data — The radiomics paper cites Haralick texture features as an early image-quantification method for extracting structured data from pixel patterns.