Robust pencil drawing generation via fast Retinex decomposition

Abstract

As imaging devices have been rapidly developed, it is very convenient to obtain natural images at all times and places. As a step further, generating the visual effect of artistic paintings from natural images has become an attractive image processing task, such as pencil drawing generation. However, current pencil drawing generation methods mainly concentrate on simulating the pencil-drawing patterns, while neglecting the complex illumination variations of an image. Therefore, various unsatisfying effects can be inevitably introduced, in turn weakening the quality of the generated pencil drawing patterns. To address this problem, we present a novel pencil drawing generation method based on Retinex decomposition. A mixed-norm Retinex image decomposition model is firstly proposed, which decomposes an image into an illumination layer and a reflectance layer. Then, with the obtained layers, the pencil line layer and tone layer are produced based on our simple but effective models, respectively. In experiments, the results show that our method consistently generates high-quality pencil drawing patterns from images with different illumination conditions. For example, as for the images with low light or back light, our method still ensures the visual quality of the obtained pencil drawings.