A Framework to Estimate Consistent Surface Color for Large-scale Outdoor Heritages

The color appearance of an object is significantly influenced by the illumination color. When illumination color changes, the color appearance of the object will change accordingly, causing its appearance to be inconsistent. Therefore, illumination condition must be set during the process of data aquisition when making a color model of an object. However, it is impossible to handle the illumination condition in outdoor environment especially when the target object is huge. To solve this, we propose a framework to correct the white balance of each aquired image taken in different time. Despite causing the inconsistency problem, the change of illumination color can produce a crucial constraint that can solve the problem itself. Finlayson et al. [10] proposed a method using this constraint. They showed that, by utilizing two different illuminations, it becomes effective to estimate a consistent surface color. Unfortunately, their method is a pixel-based operation that does not consider the presence of noise, thereby making it less robust for real-world images, particularly for those with insufficient difference of illumination colors. In this paper, we propose a more accurate and robust method that is extended from their work. We found that by examining the presence of noise and then eliminating it, we can obtain more accurate and robust results. Our implementation on those two factors using natural images shows the increase in accuracy and robustness.