Research On Line Drawings Of 3D Models

With the rapid development of computer graphics, three-dimensional computer graphics have infiltrated into every aspect of computer applications. Traditional photorealistic rendering purses the gole of realism and is usually restrained from the faithful performance of scenes and can not abandon the irrelevant details, thus can not effectively convey the characteristics of scenes. Non-photorealistic rendering (NPR) in recent years has developped rapidly, which mainly uses some sort of artistic styles to draw target objects. This NPR can show the characteristics of the objects clearly and efficiently by abstracting the target scene at different levels according to the communication expectation or mimicking a specific artistic effect on the scene. This rendering provides an effective way of expressing the scene information to make up for the limitations of photorealistic rendering methods. Line drawings, including sketches, illustrations, and comics, not only take an important part in the NPR domain, but also act as the ingrediant of other artistic styles. Line Drawing is not only able to draw a few strokes to outline the shape of models, but also has a wealth of expressiveness, artistic essential of certain emotions and feelings. Therefore, it is widely used in the areas of art creation, scientific and technical illustrations, 3D games, product design and other related domains. According to the input data type, the current research about line drawings can be devided into three-dimensional space-based rendering and image-based rendering method. The latter-the three-dimensional space of non-photorealistic rendering technique-is becoming increasingly important with the rapid development of digital media industry, such as game animation, and three-dimensional modeling techniques. Range Images, easily attained from range scanner, are very important and common sources of three-dimensional models; but most of these models are rough, noise-rich, or evern structure incompleted. Current algorithms in NPR area can not give satisfactory results on Range Images models and thus constrains the scope of Range Images-related applications.With the integration of the knowledge of art and cognitive psychology, this thesis gives a series of deep research on NPR line drawings domain, to be more specific, they are research of feature line detection algorithms based on rough, noise-rich models, typically like Range Images, research of feature line deformation, and reseach of stylistic rendering. The research aims to provide the theory and technical support for artistic and efficient rendering in the cartoon animation and other digital entertainment industry.The main contributions and creative work of this dissertation are as follows:1. Propose a novel algorithm to extract features lines from Range Images.In order to efficiently extract features from Range Images, usually with the characteristic of rough, noise-rich or even structrure incompleted, we propose an extended feature line extraction algorithm, and further chain and smooth the discreted lines into long line path which is much suitable for the following stylistic process. The main idea of this extended feature lines extraction comes from the artistic observation: most of feature lines in artistic work are not isolated, but connected with each other. Therefore, the algorithm first set a small number of reliable feature lines as the initial reliable starting lines, and then detect those feature lines in consistent with the existing feature lines as far as they satisfy the loose definition of feature lines. Therefore, this method overtakes any other previous algorithms in avoiding a large number of burr lines mainly caused by noise. This idea can also be easily applied to existing types of feature line algorithms, and thus greatly reduce the noise impact on the models.2. Propose a hybrid method based on the geometry property of 3D models and corresponding instensity images.In most cases, featue lines from 3D models seem too few to delineate the ensential of the objects. In the domain of human faces, 3D Range Image data and 2D intensity image data are complementary, when producing caricature portraits. Combining geometry lines and texture areas gives a better basis for NPR portrait drawing than a Range Image or intensity image alone could. Using a Range Image as a basis allows viewpoint dependent lines to be drawn, in a way which could not be done from an intensity image. Using a 2D intensity image allows non-geometric information, or information whose geometric boundaries are hard to determine (take the eyes, eyebrows areas in portrait as examples), to be added to the result. Clearly such an approach is applicable to other kinds of drawings than portraits. 3. Propose deformation methods awared of feature linesThe attractiveness of artistic line drawings is not only lies in its ability to fainthfully delineate the object shape, but also lies in the flexible ability of emphasis and highlight on the characteristic of target objects. Inspired by the technology of Focus+Context, the thesis proposes a feature lines deformation method which can exaggereate the most distinctive characteristic features in realtime. When intergrated with the rules from traditional cartoon drawings, the deformation enbables realtime feature exaggeration without sample training but just by simply NPR lens configuration and freely interaction. The expressive experiment result suggests that the methods can be widely used in the area of digital entertainment.4. Propose a real-time 3D portrait rendering methodThe thesis gives a method for automatically creating a caricature drawing-an exaggerated portrait-of a human face, based on simultaneous use of a Range Image (or 3D mesh) and registered photograph of the same face. While previous papers have considered producing portraits from either source individually, we are unaware of methods which combine information from these sources. Unlike most previous template-based, component-based or example-based face sketching methods, which work from a frontal photograph as input; our system uses a Range Image as input. Our method runs in real-time for models of moderate complexity, allowing the pose and drawing style to be modified interactively. We also suggest an approach to facial feature exaggeration for caricaturing a face, based on modifying it relative to an average face. While such variations from the average have been suggested before, we use a different approach to combining the exaggeration applied to different facial features.5. Propose a framework for the remote rendering of 3D models on mobile devicesMobile devices, like the Personal Digital Assistant (PDA) and smart phone, enjoy a significant growing number of users up to date. These devices have the limitations on network bandwidth, graphical resources and computing power. To overcome the disadvantages of the devices and balance the workload and performance between the client and server, we present a real time remote rendering method by transferring only the line primitives to client devices from server side, which will be rendered in the client side locally with various non-photorealistic rendering techniques. When applied into the digital museum exhibition, it allows the user to interactively go through the 3D models with his personal mobile device anytime and anywhere. Moreover, we can enrich the users' experience since NPR based images are usually eyes attracting.This research shares applications with those of traditional hand-generated techniques, including architectural and product design, technical and medical illustration, storytelling (e.g. children' s books), games, fine arts, and animation. Indeed we have solved some key and fundamental problem to some degree when dealing with lines drawing of three models on Range Images models. However, there are some aspects needed further improvements. We in the future should take more consideration of cognitive psychology and domain knowledge in order to better understanding the target object when abstraction and strulization. Moreover, we might further focus on constructing the general stylitic theory model which fundumently would push the prosperity of NPR era.