A framework for detection of linear gradient filled regions and their reconstruction for vector graphics

Abstract

Linear gradients are commonly applied in non-photographic art-work for shading and other artistic effects. It is sometimes necessary to generate a vector graphics form of raster images comprising such art-work with the expectation to obtain a simple output that is further editable, say, using any popular vector editing software. Further, this vectorization process should be automatic with minimal user intervention. We present a simple image vectorization algorithm that meets these requirements by detecting linear gradient filled regions and reconstructing the vector definition using that information. It uses a novel interval gradient optimization scheme to derive large regions of uniform gradient. We also demonstrate the technique on noisy images.