| Relighting is an important task in visualizing an object. For relighting, ones need to identify the characteristics of the object, particularly its response to an incident lighting condition. The response of the object surface to the incident lighting is characterized by its reflectance fields. This dissertation includes three relighting approaches to extracting the reflectance of the object from images while each of approaches is formulated under a different application scenario. As a result, the number of input images for each scenario is varied from many to few accordingly.;The first part of this work acquires the reflectance and geometry of an object from many images. We capture a set of many images, more than 100, of the object using our proposed system, namely a 3D/reflectance scanner. By processing these recorded images, we can obtain both the reflectance and geometry of the object simultaneously. This approach is unique in that we capture one set of images using one efficient system, which extracts both the reflectance and geometry at once.;In the second part of the work, we propose a relighting algorithm using illumination patterns. This approach is classified into image-based relighting, which processes a set of input images taken under designed illumination conditions and reconstructs the reflectance of an object for relighting. We capture a few images, more than 10, under statistically driven illumination patterns. By analyzing recorded images, we successfully reconstructed the reflectance of the object. Consequently, it is possible to generate a visually pleasing image under a new lighting condition.;Finally, we present a machine-learning approach to lighting enhancement in the third chapter. It includes a face relighting algorithm, which estimates the reflectance and lighting from as few as a single input image. This is possible by constructing a robust probabilistic reflectance model for faces. On top of this face relighting algorithm, we develop a machine-learning approach to identifying an optimal lighting and color condition for the face image. We have constructed models for the optimal lighting and color respectively using a thousand of well-photographed face images. These face images are collected from a web and captured by professional photographers. After all, we find the optimal lighting and color condition for the input image and thereby we synthesize the input face preserving a good lighting and color condition. As a consequent, the proposed lighting enhancement technique performs better than existing techniques in the subjective user study. |
