Knowledge-based Active Appearance Model Applied In Object Localization Of Medical Images

During the research of image processing and computer vision, object localization deserves much research work. Wide application has been conducted in military, medical and computer science fields, such as precise targeting weapons, computer-aided diagnosis and X-ray medical image of human spines systems. Good performance has been attained on local object localization by active appearance model (AAM), which is a statistical model of localization; however, correlations among different objects are not attached much importance. Concerning this, a knowledge-based active appearance model is brought about here. Based on the X-ray medical image of human spines, AAM is first applied to middle bone of the spine. Then correlations between this middle bone and neighbored ones are relied upon to adjust original model built on middle bones to perform localization of neighbored ones. Concerning sensitivity of AAM to starting positions, histogram-based approach is suggested to predict rough starting point. Through comparison of test results, we can easily deduce that the knowledge-based AAM brings us better performance either in time or accuracy; therefore, a good robustness has been acquired.