Is there any texture in the image?

Texture analysis methods have been used in various image processing tasks, such as image segmentation, recognition, shape analysis, texture synthesis and image compression. When applying any of these methods, we assume that the input image has some textural characteristics. This paper addresses the problem of deciding whether an image has texture; in other words, whether texture-based methods are suitable for processing the image. We define a texture to have a spatially uniform distribution of local gray-value variations. A fast algorithm for detecting regions that have texture according to this definition is presented. The performance of the method is demonstrated on several synthetic and natural images. Copyright :L~ 1996 Pattern Recognition Society. Published by Elsevier Science Ltd. Texture analysis Segmentation Low-level processing l. I N T R O D U C T I O N Texture has found wide applications in image processing. ~t'2) Examples where texture analysis methods can be used are: (i) classifying images based on their texture; (ii) segmenting an input image into regions of homogeneous texture; ~3) (iii) extracting surface-shape information from the "texture gradient"; 141 (iv) synthesizing textures that resemble natural images for various computer graphics applications; (v) retrieving images with similar textures from a database. Most methods that use texture, as well as other scene properties, such as shape, color and illuminance, assume that the image really has the desired characteristics. When applying an algorithm to a "wrong" image (e.g. segmenting a textured image by gray-level thresholding, or extracting texture gradients from a nontextured image), the result is often meaningless. In a realistic application one may not have any knowledge as to whether the objects in the image have texture or whether they can be characterized by color or shape features. With a large number of different image processing methods at our disposal, we would like to know when and how to apply these methods. In a data-driven approach the task is to automatically select an appropriate algorithm to process a given input image. This paper deals with the problem of determining if an input image has any texture and whether texturebased image processing algorithms are appropriate and will give output that makes sense for a given image. There is no unique definition of texture and each texture analysis method characterizes image texture in terms of the features it extracts from the image. * Author for correspondence. The answer to the question 'is there any texture in the image?' depends not only on the input image, but also on the goal for which the image texture is used and the textural features that are extracted from the image. This is similar to the problem of determining whether the input data has clustering tendency prior to applying a clustering algorithm to it. 151 Previous work on segmenting an image into textured and nontextured regions has focused mainly on discriminating regions of homogeneous gray value (or noise) from regions containing some structured grayvalue variations. Dinstein e t al. 16) for example, consider a pixel textured if the difference between maximum and minimum gray values in a neighborhood of the pixel is greater than a threshold. Cross and Jain (7) use Markov random field models to decide if an image region contains texture or white noise only. These two methods require that a textured region contain some gray-level variation, but not that the variation be regular. For example, areas near the edges in Fig. l(a) would be classified as containing texture. Figure 2 depicts a taxonomy representing our intuitive notion of texture. The first three classes of images in Fig. 2 (nodes labeled A, B and C) are interpreted as containing no texture. These are images of uniform gray value, white noise images and images of objects that are solely characterized by shape. Only the last three classes of images (nodes D, E and F) contain well-defined textures. The first class of non-textured images (node A) consists of sufficiently large regions of uniform color or gray value. Color and gray value, as opposed to texture, are then the pixel attributes that should be used to segment such an image. The next branch in the tree of Fig. 2 contains a random dot image (white noise) where it is impossible to define any microstructures (primitives) or spatial dependencies