-
Notifications
You must be signed in to change notification settings - Fork 0
Image Sharpening
Image sharpening, or image deblurring is the name for the process of removing motion blur from an image. When a photograph is taken in low light (hence high exposure time), with a shaky camera, or of a moving scene, the resulting image may come out blurred. This is due to the scene moving relative to the camera over the exposure time of the image.
There are four kinds of deblurring methods:
- Mechanically stabilizing the optical system within the camera.
- Recording additional data, such as camera movement, and using it to process the image.
- Deblurring from a single image with no additional data.
- Using data from a series of images to deblur.
There are a handful of well-known methods to perform deblurring on a single image without additional data from the camera: sharpening, blind deconvolution, and several newer advanced techniques - of particular interest is the method proposed at ACM SIGGRAPH 2008 by Qi Shan, Jiaya Jia, and Aseem Agartala. Some of these methods are outlined below. For more detail on each, examine the references at the bottom of the page.
Image sharpening is the most basic method of image deblurring. It is primarily used to sharpen images that were taken slightly out-of-focus, or with imperfect lenses, rather than to fix motion-blurring. However, in simple cases, it is effective at cancelling motion blur.
Sharpening takes a traditional images processing approach to the problem. The loss of sharp edges, or blurring, is interpreted in the frequency domain as a function that falls off at high frequency. The sharpening approach makes these edges clearer by passing the image through a high-pass filter, scaling the filtered image appropriately, and adding the filtered image back to the original one. There are several variations on this algorithm, involving blurring the image slightly before sharpening it, subtracting a low-pass filtered image from the original, and using different types of high-pass kernels to enhance edges differently.
Deconvolution is the main method used to deblur images. It is based on the idea that, when an image is blurred, it is essentially a crisp image with some kernel applied to it via convolution. If we know this kernel, is is straightforward to convolve our blurred image with the inverse of this kernel, producing the original crisp image. However, when this kernel is unknown, for example when it is produced by the random movement of a shaky camera and our only data is the image itself, the kernel must be estimated from the image before this deconvolution can take place. This is called blind deconvolution.
The blur kernel is more specifically called a point-spread function (PSF), which describes how light is spread from a point by an optical system. In order to estimate this from a blurry image, an obviously sharp point is taken as reference, and then the image is evaluated to find the size of the PSF. From there, there are several interative methods to estimate the PSF and subsequently deblur the image.
A big problem with this method is that in areas of high-contrast, it produces ringing. There are many methods to reduce this ringing. One of which is outlined within the Shan, Jia and Agartala method below.
Image Filtering: Noise Removal, Sharpening, Deblurring (PPT)
The Principles Behind Digital Image Sharpening
The Effect of Color Space on Image Sharpening Algorithms
Deblurring with the Blind Deconvolution Algorithm (Mathworks)
Development of Blind Image Deconvolution and its Applications
Image Motion Deblurring (Paper)
High-quality Motion Deblurring from a Single Image
Fast Motion Deblurring (Paper) -- Website
Image Deblurring Using Inertial Measurement Sensors
Coded Exposure Photography: Motion Deblurring using Fluttered Shutter