Super-Resolution Sample Clauses

Super-Resolution. In most image processing applications it is desirable to have images with high spa- tial resolution. One approach to obtain such images is to build sophisticated de- vices having intrinsically high-resolution capabilities. However, these techniques, in addition to being costly, suffer from other limitations that are difficult to over- come [94]. Super-resolution [30] is a less expensive alternative that recently has gained popularity in digital imaging and video applications. One very interesting application of this technique is in surveillance cameras, where super-resolution can help for instance in a criminal investigation. Super-resolution is an image fusion and reconstruction problem, where an im- proved resolution image is obtained from several geometrically warped, low- resolution images of the same scene. Here we assume that the geometrical warp- ing is limited to affine transformations and each of these images is shifted / ro- ▇▇▇▇▇ by subpixel displacements. The high-resolution image is not only an image that has more pixels (like in the case of interpolation), but it also has more visi- ble details. The subpixel displacements guarantee that each low-resolution image contains different information about the same scene. If the low-resolution im- ages were shifted by an integer multiple of the pixel size, then the process of super-resolution would be equivalent to an interpolation in the sense that the re- construction would not have more visible details. It should be emphasized that super-resolution is the only nonlinear inverse problem that is considered in this work; all deconvolution methods discussed above were formulated as linear in- verse problems. Suppose we have acquired m low-resolution images g(1), g(2), . . . , g(m) that meet the assumptions described in the previous paragraph.The process of super- resolution can be modeled in the following way g = K(y)f + η, (2.23) where    DS(y(1))    DS(y(2))   K(y) = .