Progressive evaluation of thermal images with segmentation and registration

• Autor(a) principal: Emilio Zorzo Barcelos
• Data de Publicação: 2015
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/BUOS-ATEKLG
• topic: Infrared thermography; Segmentation; Registration; Thermal-image analysis; Termografia; Engenharia elétrica

Several scientific applications benefit from the quali- and quantitative features of infrared thermography. However, the evaluation of thermal images is generally performed by the manual selection of regions of interest with the use of simple geometric shapes for the extraction of temperature measurements and elementary statistics. Hence, the process may be inaccurate, incomplete, expert-dependent, and time-consuming. Although the use of imaging techniques can contribute to improving the analyses, intrinsic features of infrared radiation must be considered to avoid data corruption from mishandling imagecontents representing temperature measurements. Essentially, the unavailability of a structured system of advanced image analysis methods shows to restrict the evaluation of thermal imagery to elementary statistics and lead to inaccurate interpretations because ofcorrupted data from technique misuse and imprecise selection and registration of regions of interest. This work addresses these issues presenting a new methodology for a precise and enhanced evaluation of thermal images by combining specific image processing andanalysis methods adapted for infrared thermography, including the: unassisted target detection using segmentation, regional isotherm detection, and registration of thermograms and isotherms. We introduce image segmentation for the autonomous and precise extractionof regions of interest with the use of a mask that serves as a detailed filter allowing for an accurate analysis of temperature measurements since background data is eliminated from computations. Moreover, our methodology promotes the automatic, non-rigid registrationof thermograms and isotherms by calculating transformations based on shape deformation fields that are used for aligning sequences of thermal images and isotherms to a unique spatial coordinate system without affecting the original radiometric measurements. Registration enables the extraction of comparative measures including region areasand the region-growth percentage. Experiments were conducted to evaluate the proposed methodology on sequences of thermograms taken from professional soccer players as part of a sports medicine application. Specific sets of trials portrayed the most relevant steps ofour approach with precise and consistent results. Erratic thermograms containing acquii sition artifacts and environmental distortions were selected to display the methodologys ability to cope with a broad range of issues concerning infrared thermography. The results were accurate and compatible to the manual analysis by an expert. Nonetheless, the benefit of the unsupervised segmentation and registration in the proposed methodology favors the processing of large databases, where a manual approach is unfeasible. In contrast with previous work using thermal analysis for injury assessment, our methodology is able to evaluate patterns of temperature variations over periods of time. While our method does not produce a diagnosis, its results were used by Cruzeiros medical team for assisting with the detection of potential injury and with the monitoring of treatments progress. Furthermore, because component methods have a direct functional association to the steps for the quantitative analysis of thermal images, our methodology provides a standard framework to support future work.