Automatic Design Line Detection in Urban Building Images through DCT

Abstract

The conversion of urban building images into design lines is a valuable technique used in architecture. It aids in the
analysis of design styles and building components. Recent studies have shown a significant interest in using
computational methods to examine the design lines in architectural images, such as plans, facades, and streetscapes.
This research presents an automated approach that utilizes kernel filters to extract edges and reduce noise from
various architectural images. Additionally, it introduces a new filter that separates lines based on their directions
utilizing the Discrete Cosine Transform (DCT). The proposed technique generates distinct images that display the
vertical, horizontal, and curved lines extracted from the original images. The results demonstrate that the proposed
technique is efficient not only with architectural drawings but also with photographs of existing buildings. This
technique opens the door for further experimentation in artificial intelligence, computational aesthetics, image
rectification and calibration, and 3D building reconstruction.