Exemplar-Based Photo Color Correction by Exploring Visual Aesthetics.

With the prevalence of mobile imaging devices, large amount of photos are produced in each day. Automatic image enhancing models, such as exemplar-based color correction model, are highly needed. Based on feature correspondence between the exemplars and the target photo, the model optimizes the correction parameters by solving a matrix factorization problem. However, current models do not consider how to obtain reliable exemplars. In this paper, a simple but effective idea is employed to address this issue. We introduce an aesthetics evaluation stage, which measures the quality of the exemplars, to only select aesthetically good exemplars into the color correction model. This pre-selection strategy makes the exemplars more reliable in the correction model, and thus improves the visual quality of the results. Visual and quantitative experiments validate our improved model.