A Knowledge-Based Procedure to Approach Texture Analysis

The promise of a zero failure quote (0%) through a fast and touch less inspection system [4], which can be integrated directly in the production line, has made Machine Vision Systems very popular among quality inspection and industrial automation. However, by looking in detail at the available solutions on the market, it can be notice, that most of them are based on simple image processing principles, as threshold segmentation, and the rest can only be applied for very specific tasks. Textures [1] can be defined as image region containing a pattern with specific features such as colour distribution or periodicity, e.g. a grass or wood pattern. Many tasks in the field of image processing often require texture analysis methods. Due to the characteristics of texture patterns, the processing complexity of such images is increased considerably. A general method for texture analysis has not yet been developed, so that many different approaches are used to solve very specific tasks and an uncountable variety of texture algorithms exist. These algorithms could be subdivided in statistical methods (e.g. Co-occurrence matrices, Unser and Laws texture filters), stochastic models (e.g. Autoregressive models) or frequency analysis based methods (e.g. Fourier, Gabor Filter and Wavelets). The amount of publications in the field of texture analysis is simply overwhelming. Therefore, image processing based on texture analysis has become a wide research field, which is almost impossible to be entirely overlooked. This paper proposes a knowledge-based method to manage and evaluate different texture analysis algorithms, supporting the machine vision engineer to appropriately develop an algorithmic concept for a specific application in a short time.