An Efficient Floor Plan Classification with Optimized Image Features using Machine Learning

Adding pictures is a necessary part of advertising a property for sale. Agents typically do not label images and even if they do they are not standards for such labeling. Floor plan is one of the essential categories of listing images that real estate portals would like to highlight and attract attention to. When volumes are small, manual annotation is not a problem, but there is a point where this becomes too burdensome and ultimately infeasible. Here, we propose an approach to radically increase the efficiency of such tasks. We present a novel algorithm to classify floor plan images of any type with the help of unique intrinsic image features such as mean saturation, dominant color extraction etc. and machine learning. The overall pipeline can accept any listing image, extract its features and predict whether the given image is a floor plan or not in a single shot. The overall experimentation shows that there is a performance improvement when comparing with a deep learning pipeline for the same and nearly 100% accuracy with the test data. We are also showing some additional experimentation results with other ML models apart from the best one.