Integrative Auswertung von Farbe und Textur

Integrative Analysis of Color and Texture Color and Texture are two of the most important areas of digital image processing. Both areas are facing each other without any relation. Within this work, integrative methods for analysing both phenomena are introduced. With integrative color texture features for Cooccurrence matrices and Gabor filters the statistical and the signal theoretic approach to texture analysis is covered, respectively. The separation of grayscale texture analysis and color histograms as well as the integrative view on the color texture is realized within the notation of Cooccurrence matrices.Therefore, the gain of the integrative approach is being measured by the Kolmogorov distance quantitatively. Hence, the discussion about the usefullness of the computation overhead dealing with color image processing algorithms is objectified and decided application specific. The eight most important Haralick features are used as color texture features for classification within the statistical approach. The Gabor filters are evaluated as Gabor wavelets as well as log-polar Gabor filter banks. Additionally, novel texture features are proposed. Especially the phase energy feature is shown to have similar discriminative power as the well known amplitude energy. Therefore, not only the area of color texture analysis but the entire Gabor filter method is enriched by this new feature. The Evaluation of the novel color texture features in comparison to the classic color histograms and the grayscale texture approaches is done experimentally. For this, several classification experiments are performed with the nearest-neighbor-classifier. The experiments base on three image data sets with up to 1920 images. These data cover a wide range of color and texture dominated surfaces. The examination of the experimental results shows a significant increase in classification quality by the integrative approach. Additionally, the discriminative power of intensity independent colored textures is proved. Both methods analysing colored textures are developed and evaluated for the RGB (technical) and the complex color space (perception like), respectively. The best classification results overall are seen for the complex color space. Nevertheless, no general recommendation can be given to this decorrelated color space. The comparison between both analytical methods shows significant better results for the statistical analysis by Cooccurrence matrices. Especially within the field of medical image processing, numereous application areas are open for the new approaches to colored textures for diagnostic purposes. Exemplary, this is shown within the work dealing with dermatoscopic images