Yaron Kalit (EE, Technion)
In recent years, a number of super-resolution techniques have been proposed.
Most of these techniques construct a high resolution image by either combining
several low resolution images at sub-pixel misalignments or by learning
correspondences between low and high resolution image pairs. These techniques
and most other image interpolation methods focus on grayscale images.
In this work, a new super-resolution method for color textures from a single
image is presented. The interpolation process takes advantage of the repetitive
nature of textures and the availability of several similar patches within them.
In addition, it utilizes the color-intensity correlation that often exists in
natural images, as well as local image characteristics such as smoothness and
edges in the vicinity of the interpolated pixel. The extracted information is
used in order to perform stochastic interpolation of the missing pixels, i.e.,
probability distributions are formed according to which the interpolated values
The advantage of this approach is its ability to maintain the statistical
properties in the low resolution image, as well as its suitability to a broad
class of textures. The interpolation results of the proposed method are shown
to outperform presently available methods.
Part of an MSc thesis under the supervision of Prof. Moshe Porat.