Gaussian Blur, popularized by applications like Adobe Photoshop, is a general purpose image editing filter commonly used to create drop shadows and glows. By blurring images to varying degrees, it removes fine detail and noise. The smoothing it produces leaves no artifacts or side effects.
The major difference between Gaussian Blur and other blur filters is that its effect is dependent on the level of detail within an image. As the level of detail decreases, the filter's effect decreases proportionally. Other types of blur, such as Mean Filter, do not have this feature; as a result, the variation in blur effect may change considerably. In addition to this well-defined frequency response, the Gaussian function is optimized for much faster application compared to other blur filters.
To peform the blurring effect, Gaussian Blur utilizes a type of probability curve discovered by Carl Friedrich Gauss, for whom the algorithm is named. This curve is similar to the classic bell curve, with higher probability in the center, and diminishing probability on either side. When an image is blurred in this fashion, pixels are mixed with adjacent pixels with Gaussian probability. Each respective pixel has great effect in its original location, and less effect farther away from this location, in accordance with the bell curve.
This method causes one of the Gaussian Blur's chief caveats. For each pixel, a unique convolution matrix is generated in order to model point sources with different levels of blur across the image. By this process, surrounding pixels are pulled into a point, with the side effect that the blur discs of further points will interfere with closer points. Consequently, the level of blur cannot be controlled for given pixels or regions of pixels; blur is dependent on neighbour regions of the image. However, this can usually be circumvented by using other features in image editing programs. For instance, in Photoshop, an entire image can be blurred, with the original image below in a separate layer. A gradient image mask can then be created for the top layer so that only a small area of it is visible. This, in effect, allows for a Gaussian effect on a small region of an image, with the limits of the blurred area being feathered.