Image Super Resolution By Fast Edge-Adaptive Interpolation

This paper presents a new approach to single-image super resolution by a simple and fast image interpolation method based on the combination of three different procedures. In the first phase of five phase algorithm, determination, marking and interpolation of the edges of the image is done. In second phase, the interpolated values are modified using a refinement procedure that reduces the zigzag and staircase appearance in the curved edges. In the third phase, moving average interpolation is applied to the positions of empty pixels so as to estimate the non-edge pixels with color information from surrounding neighbor pixels. The complete set of steps involved in the method results in super resolved images with a comparable appearance like that of original LR image, with much detail and least artifacts which transpire with previous methods. The colors in the various patterns are not spilled out or mixed into areas of contrasting colors, thus reducing edge blurring and providing a smooth and presentable natural-image. The proposed method shows better performance over existing approaches in PSNR value test conducted on a wide series of images in terms of computation complexity, turn-around time and quality of output. This implementation has far less computational complexities and is well suited for applications that need low turn-around time.