Feature-Based Test Oracles to Categorize Synthetic 3D and 2D Images of Blood Vessels

Automated testing activities contribute significantly to reduce the cost and to increase the productivity during the software development process. Programs with complex outputs limit the application of automated testing strategies. A possible solution is the use of feature-based oracles. In this study, we use the framework O-FIm/CO (Oracle for Images and Complex Outputs), which uses CBIR (Content-based Image Retrieval) concepts to evaluate the similarity of synthetic images of blood vessels through the "feature-based test oracle" approach. In order to demonstrate the effectiveness of the approach, we evaluated the ability and accuracy of the test oracle in automated the process of categorization of synthetic images of blood vessels in 3D and 2D models through the similarity between features. Furthermore, we compared the accuracy of the categorization of the test oracle relative to random classifiers. The results obtained in two empirical studies revealed an AVG (average) of precision, recall, and specificity of, respectively, 77%, 100%, and 88% in the categorization performed by the test oracle for 3D images and 71%, 81%, and 93% in the categorization performed by the test oracle for 2D images.