Three-Dimensional Surface Texture Segmentation Based On Probabilistic Index Maps

Texture is one of the most important features in image processing, has been applied to natural image, remote sensing satellite imagery, material surface images, medical images, etc. Texture analysis plays an important role in computer vision, image processing and computer graphics research, and it is leading a great prospect. As part of the texture analysis, 2D texture segmentation has considerably developed a complete and mature theory in recent decades. However, due to the variability of 3D surface texture on the illumination and view angles, texture segmentation methods on 3D texture cannot be well studied .Three-dimensional surface texture appearance representation depends on illumination and view angles. One 3D surface texture makes different appearances under different illumination and view angles. However, it does not change the structure attributes essentially. Since the Probabilistic Index Maps model is suitable for the texture images above, it is applied to the algorithm of 3D surface texture segmentation. This paper demonstrates the segmentation approach based on the Probabilistic Index Maps model.This thesis firstly introduces the research status of texture analysis and segmentation, and summarizes the background and theoretical knowledge of the Probabilistic Index Maps model systematically. We then briefly introduce research and application in cell recognition and division through the usage of Probabilistic Index Maps model, and prove the effectiveness of this method through experiments. We further integrate texture segmentation based on the combination of the Probabilistic Index Maps model and a fuzzy k-means clustering algorithm. This thesis proposes the Probabilistic Index Maps model based texture segmentation algorithm. Based on 2d texture segmentation, the algorithm was applied to 3D surface texture segmentation and segmentation based on the Lambertian model. In the experimental process, we made comparison from feature extractions and resolutions. The experimental results show the effectiveness of the proposed method in this thesis.