| Lines and edges are probably the most important characteristics of an image. Many image processing techniques utilize lines and edges. The contributions of this work are in three major areas of image processing and they all involve lines or edges. The first area is that of image filtering using non-linear edge preserving filters. Among them, rank-filters with median filters as a special case are the most important. Median filters (MF) are used to remove impulsive noise from images while preserving edges. In this thesis, an extension of the MF, the Vector Median Filter (VMF) is introduced. As opposed to the MF, the VMF outputs for each window position a number of data elements. Just like the MF, the VMF filters impulses while preserving edges. Its principal advantage is superior computational speed, with performance similar to that of the MF. Deterministic and statistical properties will be examined, as well as two-dimensional extensions. A fast VMF algorithm is also presented. In addition, a novel filtering scheme using MF's and linear transforms is introduced.The second area of this work is in texture analysis and classification. Texture is one of the important image characteristics and is used to identify objects or regions of interest. Texture classification techniques are either statistical or structural. In this work, a new structural approach based on line detection is introduced. This classification is based on the relative orientation and location of the lines within the texture. With proper normalization, the classification is independent of geometrical transformations such as rotation, translation and/or scaling. Associative memory encoding is used as post-processing decision maker. Iterative associative memory encoding is introduced for additional accuracy.The third area in which contributions are made is image segmentation. Image segmentation is a highly scene dependent and problem dependent decision making or pattern recognition process. Based on edge and line detection, an image segmentation technique for images containing polygonal shapes is introduced. The method is based on the fact that the sides of a polygon form a closed contour. Stray lines and noise are effectively suppressed. |
