Approach to Design Reference Management using Auto-Recognition System of Room and Design Style

Abstract

This paper aims to propose an approach to managing interior design reference images using the automatic recognition system for room and design style. In practice, architects, designers, and other interested parties actively use web-based platforms for which design references are provided with a well-organized information structure. In particular, the photograph has the role of a primary source to support analyze, and understand architectural design. However, the current management approach consumes significant time and design expertise resources. In addition, the approach to managing and labeling filenames that are irrelevant to the image itself is inefficient and can lead to errors. Therefore, we use a deep learning mechanism and pre-trained image recognition model to retrain the model with the Korean apartment room and its design style. Using image recognition technique, it is possible to classify design images with only visual pixel data and measure the visual similarity. We developed a prototype GUI application that supports checking the result of the automatic recognition and management. In addition, the reference image data can be effectively searched and utilized through more specific search functions and both stochastic keyword-based and visual similarity-based retrieval.