| Many people assume that a blurred image caused by out-of-focus optics or camera motion cannot be corrected, because the information has been lost. That's not quite true. The information is there, but has been spread out. The blurring can be described as a "convolution" of the true image with a blur function. Removing the blur to recover the original image is called a "deconvolution." Reindeer Graphics® Interactive Deconvolution is a flexible and powerful tool for using Fourier deconvolution to enhance blurred images. The plug-in generates a blur function, or point spread function (PSF), from parameters which the user can control with simple sliders or numeric entries. Interactive Deconvolution will generate PSFs for two common types of image blur - defocus blur, i.e. blurring from out of focus optics and motion blur, either from camera motion or motion in the scene.  The plug-in works for 8 and 16 bit per channel greyscale and RGB color images. No special knowledge of Fourier transforms or complex math is required, and there is no restriction of image sizes to a power of 2, as in many FFT programs. The controls are straightforward and interactive, and the preview allows users to quickly find a combination of parameters to define a useful PSF. For defocus blur, the PSF is modelled as a Gaussian function with an adjustable standard deviation. There is also an astigmatism adjustment, which can be set to any orientation. In addition to dealing with tilted lenses, this is also handy for instruments like video cameras and scanned probe microscopes that have different image resolution and blur artefacts in the fast scan (horizontal) and slow scan (vertical) directions. Good values for this parameter can easily be obtained by finding the point at which ringing artefacts appear. For motion blur, the PSF is modelled as an antialiased line, of adjustable orientation and length. This corresponds to simple motion of the camera relative to the scene, or vice versa. Images in which different regions have different motion vectors must have each region processed individually. The line length parameter can be adjusted by finding a setting that minimizes the appearance of wave-like artefacts in the result. The Interactive Deconvolution plug-in includes a noise parameter that will reduce artefacts as it is raised. Images are sharper at lower levels, but any noise in the image will grow to dominate. This parameter is ideally adjusted to the lowest value that does not generate artefacts due to noise. Too large a value produces less improvement in resolution than can potentially be achieved or blurs the image. This parameter is called the Wiener K-value. About deconvolution Deconvolution of blurred images is carried out in Fourier space by performing a (complex) division of the Fourier transform of the blurred image by that of the point spread function (PSF). The PSF shows how an imaging system degrades a perfect point, and in some applications (most notably astronomy, where a single star serves as an effective point source) it is practical to acquire the PSF directly as an image. In most cases, however, this is not possible and the PSF must be determined in other ways. In a very few cases, measurement data are available from which the PSF can be calculated. The classic example of this method is how NASA corrected the defects in the Hubble Space Telescope images before the repair mission. Another approach is to seek a useful approximation to the PSF iteratively. Programs that perform this so-called "blind deconvolution" are typically very slow, and it is hard to find a good stopping point for any interative method. Iterative blind deconvolution depends critically on the user-defined boundary conditions. These may include such things as non-negativity (i.e., that no pixel can have a meaningless negative value) or that most of the image should be dark with only a few bright points (this works in both astronomy and fluorescence microscopy). Still another way to obtain a PSF for a system is to capture an image of a known test pattern (preferably one with detail and sharp edges in many orientations). Deconvolving the image in question with the known pattern gives the PSF. All of these methods require a priori information about the system and/or the image, which is often difficult or impossible to obtain. Interactive Deconvolution allows the user to create a PSF and observe the effect of deconvolution in real time. The plug-in uses a classic method known as Wiener deconvolution. Mathematically, this provides a well-behaved method for preventing the Fourier division from enhancing noise that could corrupt the resulting image. Weiner deconvolution Fovea Pro includes a powerful classic deconvolution plug-in to be used with a measured or derived PSF using apodization to avoid numeric overflow, and can be used to extract a PSF from a blurred image of a known pattern. When you cannot capture the actual PSF, the interactive method allows users to apply their knowledge and judgment in a particularly straightforward way. 
Example with regular deconvolution (when you already have the PSF). The Weiner-K Noise Level is used to adjust the sharpness of the result.
Next: Summary of Contents
Prev: Automation
Up: Fovea Pro | 