Self-example based super-resolution with fractal-based gradient enhancement

Recently, the example-based super-resolution method has been extensively studied due to its vivid perception. However, this kind of method directly transfers the high-frequency details of the examples to the low-resolution image, incurring false structures and over-sharpness around the texture regions. In this paper, the problem in the example-based method is investigated from an analytic discussion. Then we propose a super-resolution method that reconstructs sharp edges using the redundancy properties. The super-resolution problem is formulated as a unified regularization scheme which adaptively emphasizes the importance of high-frequency residuals in structural examples and scale invariant fractal property in textural regions. The experimental results show that the high-lights of our method exist in the enhanced visual quality with sharp edges, natural textures and few artifacts.