Machine Learning-based Binarization Technique of Hand-drawn Floor Plans

Abstract

Purpose: In this study, we propose a two-step binarization method for hand-drawn architectural floor plans to transform them into usable formats for indoor spatial modeling.Methods: First, a Gaussian mixture modeling was adopted to remove texture-like noise from the background. Second, 24 features were extracted to train the random forest model and the remaining line or spot-like noise was removed from the image. Moreover, the proposed method was applied to a completely different architectural drawing set to evaluate its generalization performance.Results: The experimental results indicated that the proposed method outperformed the other binarization techniques. Moreover, binarization result was outperforming with 0.987 F1-score. Conclusion: The experimental results showed that the overall performance of the proposed method was significantly superior to that of the other binarization methods. Moreover, they indicated that the proposed method is applicable to different types of architectural drawing, thereby proving its generalization.