Image deblurring for satellite imagery using small-support-regularized deconvolution

This paper presents a new practical deblurring method, small-support-regularized (SSR) deconvolution, for low quality remotely sensed imagery. In the case that the causes of image blur are various and complicated, a Gaussian degradation model is employed to approximate the composite effect. The model in the frequency domain is deduced which yields a representation with the same small support as the Point Spread Function (PSF). An approximate regularized deconvolution filter is proposed. The regularization term of the deconvolution filter is defined as a function relevant to the equivalent image power spectrum. All the computations to derive the deconvolution filter are implemented in the same support as the PSF. By this method, large matrix manipulation is avoided and remote sensing images can be filtered one at a time, without memory limitations. Meanwhile the method increases computational efficiency, which is most important for large scale satellite images. A case study was conducted for a Chinese small earth observation satellite HJ imagery. The deblurring result proves that this method successfully restores fine image detail, particularly for line features. Various measurements of the image quality show that the algorithm is comparable with other state-of-the-art methods and has advantages for image contrast and edge strength. The computation efficiency increases by about 8–37% for images with sample sizes from 256 to 1000, and will increase more for larger image sizes.