On sensitivity of texture parameters to smooth variations of local image intensity and contrast

Nonuniformities in illumination of a scene and/or in imaging device response can deteriorate texture analysis results since some texture parameters depend not only on the properties of visualized structure but also on local mean and variance of the image. This work is an attempt to discuss the sensitivity of texture parameters to image inhomogeneities (nonuniformities) - to select those which do not alter their values significantly. The properties of widely used texture parameters, i.e. based on gray level co-occurrence matrix, run length matrix and wavelet transform are taken into account. Two nonuniformity models are considered: smooth variations of image intensity and of its contrast, respectively. The discussion is illustrated by classification examples performed for natural textures from the Brodatz album. In conclusion, some features are identified, which represent certain immunity to the discussed image inhomogeneities.