Fast and Robust Building Extraction Based on HSV Color Analysis Using Color Segmentation and GrabCut

Abstract

In this paper, we propose a method for automatically extracting buildings from scenery images. The method utilizes color segmentation and GrabCut on the basis of the fact that the background regions of scenery images tend to be found in the upper and lower parts of the images. We evaluated its extraction accuracy and computational time by using 106 high-quality scenery images (HQ dataset) and 89 low-quality ones (LQ dataset). Experiments showed that k-means clustering for color segmentation and HSV color space allow the proposed method to achieve higher extraction accuracy and faster computational time compared with the conventional method. The proposed method improved the extraction accuracy by 14% or more and reduced the computational time by 5% or more for both datasets compared with the conventional method. Comparing the extraction accuracy of the proposed method by using different color spaces, HSV color space improved the accuracy by more than 2.78% for the LQ dataset due to its noise robustness. The experiments, however, suggest that the proposed method has room for improvement in terms of the process of generating the initial seed used to initialize GrabCut.

Keywords

• scenery image
• building extraction
• grabcut
• color segmentation
• building recognition