Featured-based body registration using outer contour and inner frames from thermographic images

Converting pixel coordinates from each anatomical position to a uniform shape is an important task in functional thermograph diagnosis. For the region of interest (ROI), body shape is standardized for temperature analysis and parametric computation. In this study, a four-stage automatic registration protocol was proposed to convert human body shapes from patient space to uniform standard space. First, grayscale binarization and continuous erosion were applied to segment body shape to obtain the outer contour and inner frames respectively. Second, the feature/corner point extractions were then performed to eliminate the majority of pixels. After that, the corner point matching procedure was utilized to delete asymmetric corner points between patient and standard spaces. Lastly, the Beier-Neely metamorphosis was used to transform pixels inside outer contour to their correct anatomical coordinates in standard space. The appropriateness of the proposed algorithm was evaluated by six binarized body shapes with different posture misalignments. The mapping error of background and foreground were less than 0.8% and 4% respectively. In addition, converted images can reflect the corresponding anatomical relationship between thermographs and human body. The four-stage registration method was proven to achieve acceptable mapping performance. The temperature distribution of populations can be compared on the basis of uniform shape. And the detected abnormal regions by bio-statistics technique may be useful in assisting digital diagnosis of thermographs.