gradientblur

gradientblur is a digital image processing technique in which the blur applied to an image varies across the image according to a gradient map or image gradient. The effect can simulate depth of field, motion blur, or artistic diffusion by applying stronger blur where the gradient indicates out-of-focus regions or motion.

Implementation usually uses a spatially varying blur kernel: for each pixel, a local blur radius is computed from a gradient value, and a convolution is performed with a corresponding kernel. Practical implementations often approximate this with: (a) creating a blurred version of the image and blending it with the original using a gradient mask; (b) using a multi-pass separable blur with a variable radius; (c) GPU shader approaches that sample the image multiple times with radius tuned by a gradient texture.

Common gradient sources include depth maps from stereo or RGB-D sensors, luminance/contrast gradients, or explicit user-defined masks. Variants include linear gradients along an axis, radial gradients centered on a point, or arbitrary gradient maps. Applications include simulating depth of field in photography, focus falloff in visual effects, and stylized blur effects in graphic design.

Challenges include maintaining edge fidelity to avoid halos, handling high-radius regions efficiently, and ensuring consistent performance on large images. Techniques such as edge-aware masks, bilateral-weighted blends, or segmentation masks can help preserve content boundaries while applying gradient blur.

See also: blur, depth of field, selective blur, variable-radius blur, convolution, shader.