Geometric Locally Adaptive Sharpening Method
Tech ID: 21883 / UC Case 2011-864-0
Blur and noise are the two common problems that exist in digital imaging. An important camera setting that strongly effects these two distortions, and that needs to be carefully adjusted, is the aperture size. If the exposure time is fixed, a large aperture will increase the signal to noise ratio (SNR), meanwhile reducing the depth of field (DOF) and thus increasing the out-of-focus blur, which eliminates high-frequency components of the image. On the other hand, a small aperture will alleviate the blur but increase the noise level (digital equivalent of film grain). Noise can also be suppressed by using longer exposure time; but of course, this may cause motion blur that is even more difficult to remove. At the same time limited accuracy of auto-focus systems and low light condition may add extra blur and noise into the image. So in real applications, such as consumer digital imaging, it is very common to record weakly blurred and relatively noisy images.
UCSC researchers have developed a locally adaptive restoration algorithm based on an extension of the steering kernel regression technique. This non-parametric approach is called geometric locally adaptive sharpening (GLAS). It is able to capture local image structure and thus effectively combine denoising and sharpening together without either noise magnification or over-sharpening artifacts. It is also capable of removing chrominance artifacts. Experiments show that the new approach can efficiently restore images distorted by weak blur and strong noise compared with other state of the art adaptive sharpening methods.
ApplicationsConsumer digital imaging
This algorithm can do image sharpening and denoising simultaneously without either noise magnification or over-sharpening artifacts. It can also remove chrominance artifacts caused by, for example, demosaicing process.
Camera, noise, aperture, depth of field, signal to noise ratio, image, focus, blurred, steering kernal, GLAS, sharpening, non-parametric, geometric, chrominance artifacts, Adaptive sharpening, defocus blur, denoising, kernel regression