Research On Artistic Style Generation For Random-needle Embroidery

Artistic style generation is a new topic in computer graphics and image processing.It aims to simulate the styles of different arts by developing various rathematical models and algorithms with the aid of computer,which have been intensively used in many fields such as film production,animation design,game entertainment,cultural education and digital cultural protection.However,because of the increasing diversity and complexity in styles,the creation of entirely new and original forms of arts is still one the toughest challenges in artistic style generation.In this thesis,with the application background of inheritance and protection for a newly moder Chinese embroidery art-Random-needle Embroidery,for the first time we present an artistic style generation framework to simulate its style according to its characteristics of vector and pixel space dependency,stitch pattern flexibility,and creation scale variability,and research on several techniques in this framework,such as style primitive definition and generation,visual content abstraction,and content mapping and style synthesizing.Regarding these issues,this thesis engages in the following work:(1)We propose a vector and pixel space combined stitch definition and generation method.This method utilizes the"divide and rule" strategy by dividing the complicated thread distributions of Random-needle Em-broidery into various style primitives called "stitch",i.e.,collections of threads arranged in different patterns.The stitch is defined and generated inseparably both in vector and pixel space.In vector space,we present a stitch model which can define and describe complex thread distributions of stitches from orientation uniformity,directionality,average length,density and width.In pixel space,based on different thread distributions of stitches,we propose a series of rendering method including scan-line algorithm,multi-kermel image convolution,image sharpening,etc.to generate stitches with various styles on 2D pixel arrays.We also organize the generated stitches according to their styles and generate a stitch dictionary to give a compact representation of the stitches.Accordingly,we construct the relation between free-form threads and their visual appearances.(2)We propose a visual content abstraction method based on Voronoi Diagram and color subset selection for Random-needle Embroidery.This method takes the digital image as input,and based on adopting hierarchieal image segmentation and multi-feature fusion image parsing method to extract regions and their visual information,Central Voronoi Diagram is used to partition each region into sub-regions with different size and orientations,and relevant visual information are integrated to describe each sub-region,which realizes region decomposition and representation under the need of local detail and orientation control.Then,a color quantization method based on color entropy sorting is proposed to find a color subset from a specific color library,which can describe the colors for each subregion.Together,we construct a coarse-to-fine structure on the basis of object regions and their visual information to describe the input image,which satisfies the requirements of dynamic visual content description for random-needle style generation and can guild the subsequence stitch selection and layout.(3)We propose a content mapping and style synthesizing method based on sparse representation and graph cuts optimization for Random-needle Embroidery.This method utilizes stitch as medium to realize the mapping from source image content to the style of random-needles.By introducing sparse representation into this mapping process,various image content can be conveyed through adaptively selecting the optimal linear combination of the stitch dictionary atoms,which realize utilizing limited predefined stitches to fit continuous image contents.During style synthesizing process,texture synthesis based on graph cuts optimization is adopted to find an optimal seam in the adjacent stitches' overlapping area.Thread continuity constraint is also incorporated into this process to ensure the continuity of the texture,which eliminates the artifacts in the result image.Thus,by increasing the adaptability of content mapping process and naturality of style synthesizing,it adapts to the characteristics of utilizing various stitches to reproduce image content in Random-needle Embroidery,and can generate style image without any artifacts.