Abstract: Repeated texture is one of the most important features in building facades. How to automatically detect repeating textures from facade images is an important part of building facade analysis. This paper proposes a new Baseline Constrained Texture Detection Method to automatically detect the exact location and size of repeating objects. The method first uses Bayesian adaptive superpixel segmentation to construct a superpixel adjacency map, and then separates the wall by calculating chromatic aberration to obtain candidate objects. Secondly, extract straight lines from the original image, and perform preprocessing such as normal classification and line segment clustering on the obtained straight lines. Finally, texture repair is performed based on prior knowledge of building facade structure. The experimental results show that the proposed method can effectively detect the position and size information of repeated textures when detecting repeated textures whose geometric shapes are rectangles, and repair the occluded repeated objects.

Key words: superpixel, chromatic aberration, baseline constraint, repeated texture detection, texture repair