Learning deep representation using a fully convolutional autoencoder for automated floor plan image retrieval

The real estate industry is going digital by transforming its entire project workflow, such as planning, modeling, and designing. One of the pivotal requirements of the early design stage of any infrastructure design is coming up with a floor plan. For an experienced architect also, referring to the existing model to get inspirations and ideas is very common. Searching through a database of floor plan images manually and identifying the plan of interest is a cumbersome process. This chapter presents a system for the retrieval of similar floor plan images from a large-scale database under the query-by-example paradigm that is proposed. The proposed method uses deep representation learning to extract essential features from the floor plan images. These features are then used for similarity measurement for image retrieval. We contribute to the research in two ways. First, by proposing a new query-by-example-based floor plan retrieval framework for a large-scale dataset. Second, a fully convolutional auto-encoder architecture is proposed for discriminative feature representation and fast retrieval. An experimental study, followed by qualitative and quantitative comparison with state-of-the-art methods, authenticates our proposed floor plan image retrieval system.