Suppression of inhomogeneities in images of textured surfaces

Abstract

Automated visual inspection tasks are frequently concerned with the examination of homogeneously textured surfaces such as fab- rics, wallpapers, machined surfaces, and floorcoverings. Often, the im- ages taken from such surfaces are degraded by an intensity inhomoge- neity due to the image acquisition process. This inhomogeneity is considered to be an irrelevant and disturbing signal component, which should be suppressed to enhance the desired texture component and to ease a subsequent texture analysis. We show that, especially for tex- tured surfaces, it is not always reasonable to assume a pure multiplica- tive composition of the texture signal and a disturbing inhomogeneity. We introduce a notion of homogeneity of n'th degree based on first-order statistics and present image processing methods for the homogenization of first, second, and infinite degree. For the homogenization of second degree, we propose a computationally efficient frequency domain signal processing method with high homogenization performance and low non- linear distortion. Furthermore, we suggest a high-performance homo- genization of the infinite-degree technique that equates the local histo- grams to a global histogram, which is adapted to the given image data. We compare the proposed homogenization methods visually and quan- titatively with the well-known homomorphic filtering technique, which as- sumes a pure multiplicative inhomogeneity. We demonstrate that our methods achieve much better results for synthetic as well as for realistic images of textured surfaces. © 1997 Society of Photo-Optical Instrumentation En- gineers.