This thesis introduces unsupervised image analysis algorithms for the segmentation of several types of images, with an emphasis on proteomics and medical images. Segmentation is a challenging task in computer vision with essential applications in biomedical engineering, remote sensing, robotics and automation. Typically, the target region is separated from the rest of image regions utilizing defining features including intensity, texture, color or motion cues. In this light, multiple segments are generated and the selection of the most significant segments becomes a controversial decision as it highly hinges on heuristic considerations. Moreover, the separation of the target regions is impeded by several daunting factors such as: background clutter, the presence of noise and artifacts as well as occlusions on multiple target regions. This thesis focuses on image segmentation using deformable models and specifically region-based Active Contours (ACs) because of their strong mathematical foundation and their appealing properties.ACs are formulated according to an energy functional defined so as to be minimized when approximating target boundaries. The argument of the energy functional is typically a curve or surface, which evolves and defines the partitioning of the image based on external forces that hinge on image features such as intensity and/or texture. Additionally, internal constraints generate tension and stiffness, which preserve the smoothness and continuity of the model by preventing the formation of sharp corners. The corresponding Euler-Lagrange equation constitutes a Partial Differential Equation (PDE), i.e. an iterative gradient descent algorithm, which guides the evolution towards the minimum. The numerical implementation of the evolution is performed by the level set method, which endows the model with topological adaptability, i.e. splitting or merging, appearing or disappearing during the surface evolution. In this thesis, two different objectives are pursued. The first is the core issue of unsupervised parameterization in image segmentation, whereas the second is the formulation of a complete model for the segmentation of proteomics images, which is the first to exploit the appealing attributes of ACs.The first major contribution of this thesis is a novel framework for the automated adjustment of region-based AC regularization and data fidelity parameters based on local image geometry information. Very often, AC parameters are empirically adjusted on a trial and error basis, a process which is laborious and time-consuming, based on subjectivity. On one hand, non-expert users such as Medical Doctors (MDs) and biologists require technical support since they are not familiar with the algorithmic inner mechanisms. On the other hand, parameter configurations empirically determined by image analysis experts are usually suboptimal and applicable to Correspondence to: Recommended for acceptance by Jorge Bernal
