Search

Delayed Enhancement Magnetic Resonance Imaging can be used to non-invasively differentiate viable from non-viable myocardium within the Left Ventricle in patients suffering from myocardial diseases. Automated segmentation of scarified tissue can be used to accurately quantify the percentage of myocardium affected. This paper presents a method for cardiac scar detection and segmentation based on supervised learning and level set segmentation. First, a model of the appearance of scar tissue is trained using a Support Vector Machines classifier on image-derived descriptors. Based on the areas detected by the classifier, an accurate segmentation is performed using a segmentation method based on level sets.

In the field of Cultural Heritage, image analysis represents an indispensable practice for restorers and Institutions called to plan restoration interventions. Quantification of the extension of degraded areas is usually performed by manually tracing the decay zones. This procedure is cumbersome and time consuming; in addition it is subjective and requires expertise. Moreover, the recent development and widespread of technologies able to exploit different characteristics of light (i.e. infrared and x-ray spectroscopy) and therefore provide images of phenomena that are invisible to the naked eyes suggest applications that could be improved by automatic analysis in different fields of investigations. In addition to these aspects, as remote sensing technologies nowadays allow to provide reality-based models of artefacts, the possibility to automatically extract and easily manage information derived from 2d images in a 3d environment could enrich documentation about their state of preservation. The purpose of this contribution is to show the results of investigations held in order to provide a methodology for the automatic detection of decay areas within architectures and artefacts using colour images as a field of examination. Within our investigations, we selected images representing recurrent decays, as, for example, detachments, cracks and chromatic alterations and run them both to manual and to automatic recognition and selection tests, in order to subsequently compare the results obtained using the two approaches and evaluate the reliability of the automatic one. In particular, automatic detection was based on the analysis of the histograms of the three components of rgb images from which automatic selection of thresholding values were computed and morphological operators were applied. Following this analysis, a regularization step based on level set techniques was applied to optimize the detection of decays areas. Results comparison included computational and user time, quantification of the decay area error between manual and automatically detected zones in both percentage and overlapping terms as well as the Hausdorff distance between the detected contours. Additional parameters characterizing the type of decay (i.e. crack length, mean crack width, type of detachment) were also computed for each case study. Automatic analysis in all case studies resulted faster than the manual analysis (mean time: 19 sec vs 9 min). Mean area difference between the manual and automatic analysis was 0,66%; non overlapping area resulted in a mean value of 0,42. Mean Hausdorff distance was 1,8 cm. An example of the qualitative comparison of detected missing bricks is shown in figure 1 together with the segmentation of the 3d model using the automatic detection on 2d image as a source of selection of polygonal faces. Figure 2 shows the results obtained with automatic segmentation and decay completion of a frescoed wall. Comparison between the automatic and the manual procedure showed that the automatic detection is faster and reliable in all our selected case studies. In addition, our methodology showed evident improvements in the segmentation of reality-based models derived from remote sensing, with important consequences in the evaluation of the entity and extension of decay areas on 3d geometry.