Tetyana Ivanovska

• Segmentation is an important tool in image processing. The principal goal of the segmentation process is to separate the structures of interest from the background and from each other. The focus of the current work is to explore and develop a range of efficient segmentation approaches for multichannel image data. As main examples of application, we use synthetic color images as well as medical and biomedical image data. In our studies the objective is to divide the entire image into subregions. Such a procedure is often called partitioning. We start from a generalization of grayscale segmentation techniques to multichannel data by converting it to a scalar field. We propose a procedure that converts color to scalar data while preserving as many salient features as possible which is important for segmentation purposes. We apply standard segmentation techniques to the converted data and discuss the advantages and limitations of such conversion. Apart from that, we propose an approach that allows for the direct segmentation and surface extraction mechanisms for color data. Our approach consists of several automated steps and an intuitive mechanism for user-guided feature selection. Often operating only in the feature space is not fully adequate, and algorithms which operate both in feature and object space are preferable. We analyze several global multichannel image partitioning approaches using minimum description length. We develop a novel extension of the approach to multichannel image partitioning. In certain highly specialized segmentation tasks, prior knowledge, e. g., feature and shape characteristics, about the areas to be extracted might be useful. For such problems a relatively simple partitioning can be used in combination with several post-processing steps. We apply such approach for evaluation of certain cell types. We discuss the results and limitations of the individual processing steps as well as of the overall automatic quantification approach.