Spline snakes for object recognition in medical and material images

The main objective of this thesis is to detect the contours of objects or their portions in 2D digital images using various methods that will be compared. We mainly focus on two applications, which correspond to biomedical imaging techniques and particle detection in microscopy images. In particular, we address the detection of tumour lesions within mammography images and the quantitative analysis of microplastics acquired with optical microscopy techniques. Regarding the medical field, the first application is the detection of lesions within mammograms. Then, we implement an algorithm based on image segmentation to calculate the density percentage of breast from mammography images. The knowledge of this percentage is closely linked to the BI-RADS classification of the American College of Radiology and it is based on the distribution of the fibro-grandular tissue (the densest and more radio-opaque) with respect to the adipose tissue. In the field of materials science, we detect the contours of individual micro-particles in a 2D image acquired with an optical microscope analysing a sample of microplastics dispersed in water. Furthermore, we characterize the sample composition, in terms of particle size, shape and presence of aggregates, in order to provide a quantification at single-particle level. In summary, some theoretical definitions concerning the concept of digital image and digital image processing are given in the first chapters focusing on morphological operations, image segmentation and edge detection. Then, we explore the application of various image segmentation techniques like those based on the first derivative, histogram-based thresholding of B-splines, active contour snakes, Gradient Vector Flow (GVF) snakes method, Generalized Gradient Vector Flow snakes method (GGVF) and Image Structure Adaptive Gradient Vector Flow method (ISAGVF). For the detection of microplastics we also investigate the application of the technique of watershed to particle aggregates for a more refined segmentation, after assessing the impossibility of doing this task by parametric analysis alone.