Stylized Rendering Based On A Single Image

Non-photorealistic rendering aims to synthesize images with specific style in our real life.We studied and proposed rendering frameworks for two stylish rendering methods based on a single image,cartoon face caricature rendering and low-poly style rendering.Cartoon face caricature rendering is divided into many different types.We focus on the political cartoons,which use simple lines to simulate the characteristics of the figure and exaggerate them to provide a more amusing effect.As for low-poly rendering,we use triangles with different colors and sizes to abstractly represent the reference image.The difficulty mainly lies in reproduce the content of the image with limited amount of triangles,and at the main time,to ensure that the results is artistic and rendering time is real-time.First of all,we analysize face region and extract features from a single image.And then,binarize the face components to get cartoon-rendered faces of simple lines,and exaggerate them to make an amusing effect.Finally,we render the hair into the same style.As for low-poly rendering,we propose that it is an effective way to preserve information in the original image by taking advantage of the edges in the image.In the meanwhile,we proposed a video processing method to effectively eliminate the jittering effect.The main contributions of this paper are:(1)Proposing a layered processing framework for face components and face contour,which uses Active Shape Models to calculate landmark positions as masks and thus eliminate the interference of each other;Proposing to use Adaptive Thresholding to binarize face components to get a better face cartoon rendering result and Effectively ease the interference caused by light distribution;(2)Proposing a hair rendering framework that renders hair into the similar style like face components,and enhances and strokes the hair afterwards;Implementing and comparing two face exaggeration methods,calculating face features correctly and exaggerating correspondingly to produce a more amusing rendering result;(3)Proposing an image and video processing framework for low-poly style rendering,which largely decreases the rendering time than previous methods,and ensures an artistic result that preserves most information in the input image;Proposing to select pixels more frequently on the edges of the image to form triangles,so that the input image can be presented with limited amount of triangles;Proposing a method for images containing faces,which uses ASM landmarks to preserve face components and other important information;Proposing to fill each triangle with its color at the center of mess,and this constant-time method makes rendering more efficient;Proposing a video processing improving method,effectively eliminating the jittering effect caused by the abrupt changes in triangle positions in two adjacent frames;(4)Evaluating the two frameworks through objective and subjective tests and comparison with manually drawn results by artists,and assessing the effectiveness of the algorithm.