Several important problems in color imaging can be traced to differences in how cameras and humans sample the spectral properties of light. Color processing within the imaging pipeline, loosely referred to as color correction, transforms the sampled camera responses to a form that matches the human responses. The accuracy of the color correction transformation is limited for two reasons. First, the human visual system and most color acquisition devices critically undersample the spectral information, making the differences in their sampling functions quite significant. Second, the human visual system derives a relatively constant surface color appearance despite variations in the illuminant, complicating color correction with the need to estimate the illuminant.; Assuming complete knowledge of the illuminant, we formulate color correction as an input-referred estimation problem. In particular, we analyze how a small number of camera measurements can be used to estimate a complete spectral surface reflectance function. We introduce conventional linear color transformations, and then extend these transformations using forms of local linear regression that we refer to as submanifold estimation methods. These methods are based on the observation that for many data sets the deviations between the signal and the linear estimate is systematic; submanifold methods incorporate knowledge of these systematic deviations to improve upon linear estimation methods. We describe the geometric intuition of these methods and evaluate the submanifold method on printed material data and hyperspectral image data.; Next, we discard the assumption of complete knowledge of the illuminant and analyze a technique to estimate the illuminant. Conventional algorithms rely on statistical assumptions about the scene properties (surface reflectance functions and geometry) to estimate the ambient illuminant. We introduce a new illuminant estimation paradigm that uses an active imaging method to measure scene properties, ending the need to make significant assumptions about these properties. The method actively emits light into the scene and estimates the surface reflectances from a conventional image and an auxiliary image acquired with a flash; the ambient illuminant spectral composition is classified, in turn, using these estimated reflectance functions. We evaluate the method's stability with respect to changes in the scene statistics.
展开▼