Image super-resolution reconstruction by sparse decomposition and scale-invariant feature retrieval in micro-UAV stereo vision

To meet the need of micro-UAV vision system for navigation task, this paper introduces a new single image super resolution approach, based on the sparse decomposition and wavelet transform scale invariant features of captured images. Through the training of high resolution natural image and the blurred and down sampled low resolution image, the approach learns the overcomplete dictionaries that could represent their respective images sparsely. Based on the sparse representation at low resolution and the wavelet transform scale invariant features, the approach could automatically retrieves the high resolution patches for the super resolution reconstruction. Compared with the celebrated algorithms proposed in the literature, this approach improves both PSNR of the recovered image and appearance which would improve the stereo vision for scene depth estimat ion.