Software Feature Detection using Source Code Minimaps as Visual Signatures

This study introduces a novel approach for software feature detection using source code minimaps as visual signatures. Although software analysis tools exist, to our knowledge, no work leverages visual abstractions such as minimaps combined with texture analysis for source code stereotype detection. These minimaps provide a compact visual abstraction of source code, encoding its structural and textual patterns, which are then analyzed using texture-based techniques. The proposed method employs Local Binary Patterns (LBP) to extract texture features from minimaps. It utilizes machine learning models, including K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machines (SVM), to classify software components into predefined stereotypes. A custom dataset was created from real-world software projects, including fixed-size and variable-size minimap representations, enabling a comparative evaluation of the method. Experimental results demonstrate the approach’s efficacy, with classification accuracies reaching up to 93% for RF. The findings highlight the potential of source code minimaps combined with machine learning (ML) as a scalable solution for software analysis, opening pathways for further exploration in feature detection and classification in diverse programming environments.