Relighting Multiple Color Textures
Diao Chang-yu,Lu Dong-ming,Liu Gang
DOI: https://doi.org/10.1631/jzus.2005.a1284
2005-01-01
Abstract:With the development of digital library technology, library books made of paper can be digital released and read, and Endangered Cultural Heritages can be preserved. Traditional library's contents and functions can be greatly enhanced by digital technologies. For these new library objects, the primary key problem is precisely reconstructing their 3D models. When constructing complete 3D models, multiple color, texture maps are often necessary. A commonly encountered problem uncounted during fusing of textures from multiple color images is color distortion. Each texture of a single 3D model may be obtained under possibly different lighting conditions and color response of the camera. To remove any visible seam and improve color consistency between the textures while avoiding color distortion, we propose a new efficient algorithm to relight all the texture images globally, spread residual light difference, and recolor each image by homogeneous transformation. A relative illumination model was adopted to obtain the relighting function. We chooselaß color space with minimal correlation between channels for many natural scenes, for calculating the relighting result. Looking into two overlapped imagesA andB, we can pairwise relightB intoA's luminosity condition in two steps. We first scaleB'sl channel by thelA/lB ratio of the overlapped region. We can assumeA andB are in a same color plane now. Then a homogeneous transformation, is applied toB's α and β channels which movesB intoA's hue and saturation condition. For multiple overlapped color textures, a patch based weighted, global relighting method was proposed to minimize the total color difference. The pairwise relighting method was used between each two overlapped images, and the difference in every overlapped region after, relighting was weighted and summed up to construct an energy value. We used Nelder-Mead method to find a minimal energy value and the relighting parameters for every image. After global relighting, textures become almost coherent. We simply blended the overlapped region along the texture border to remove small visual seams and get a final result. We illustrate our method by calibrating textures of a painted sculpture acquired with laser scanner. Experimental results were realistic and reliable and showed how this method can fuse multiple textures without color distortion.