Vector Image Generation And Manipulation Based On Diffusion Curves

Vector graphics use geometric primitives such as points,lines,curves,and polygon to represent image content.It is resolution-independent,more compact,and easy for editing.It has a wide range of applications in creation and other fields.Diffusion Curves is a curve-based vector representation.It uses cubic Bézier curves and attached both side color control points to create high-quality vector images.Our research focus on the generation and editing of Diffusion Curves.To generate diffusion curves image,users can create curves with color attributes manually or extract geometric and color information from existing bitmaps.The manual creation method relies on the skills of the creator,so we usually create it by image vectorization.The method based on image vectorization relies on the accurate extraction of object edges in the image,and it is easy to miss the edge of the image in the place where the color gradient is small.With the popularity of depth cameras,images with depth information are becoming easier to acquire.We propose a method for image vectorization to generate Diffusion Curves,which can be applied to RGB and RGB-D image vectorization.We use multi-scale Canny edge detection to extract edge from RGB image,and use the depth map to assist in edge extraction for the image with depth information.Color and blur information are extracted from color image according to the edge information.As a result,our method can generate diffusion curves with higher quality.To edit diffusion curves image,because the geometric primitives of Diffusion Curves are discrete and lack topology constraints between each other,the existing image editing methods are limited to local detail modification based on a single curve,and it is difficult to operate on the overall image content.Our research proposes an editing method of diffusion curves,which can be used to perform local image editing and global image editing.For global editing,we discretize the Bézier curves and perform space deformation on the image by linear blend skinning,then refit the deformed curves to Bézier curves.Our method allows users to edit the global shape of diffusion curves image while preserves its data structure for later operations.In order to reduce editing interactions,we add the continuity constraints between connected curves to preserve the local details.Combining local and global editing operation,users can edit diffusion curves more flexibly and efficiently.