2020 Volume 39 Issue 6 Pages 275-290
In general, medical records such as surgery consent forms include body part diagrams where doctors illustrate the positions, sizes, and conditions of disease. The information on which kinds of body part diagrams are in medical records is important and useful for the management and knowledge discovery of medical records. This study aims to propose and develop a method that automatically recognizes the kinds of body part diagrams. The past study applied a method consisting of the modified weighted direction index histogram and the modified quadratic discriminant function, and resulted in a recognition rate of 98.52 ［％］. However, there is room for performance improvement, and the processes and settings of the conventional method were complex and difficult to configure. The present study aims at better performance and simpler and easier processes and settings. We propose a method consisting of histograms of oriented gradients and a Mahalanobis-distance-based classifier, and conduct evaluation experiments for the recognition of 10 and 30 kinds of body part diagrams. The experimental results show that the proposed method achieves 100.00 ［％］ generalized recognition rates under almost all conditions, and is superior to the conventional method.
