Kernel‐Based Texture in Remote Sensing Image Classification

Abstract Texture has been of great interest to remote sensing analysts for more than three decades. This paper is a review of texture approaches that are based on a moving window, or kernel, and that generate a summary measure of local spatial variation, which is assigned to the central pixel of the kernel. Texture methods are challenging to implement, partly because of the many parameters that need to be set prior to running a texture analysis. The list of parameters includes the texture order, metric, kernel size, and spectral band. For second‐order metrics, additional parameters that need to be set include radiometric re‐quantization, displacement, and angle. Although few general rules of thumb can be provided in selecting texture analysis parameters, understanding the conceptual role of these parameters helps illuminate the options available. In addition, future opportunities in object‐oriented texture, adaptive texture measures, and multi‐scale texture fusion offer the potential for addressing some of the inherent challenges in the application of texture in image analysis.