Contributions To Several Issues Of Procedural Modeling Techniques

Interaction, immersive visualization and imagination are the three objectives in the Virtual Reality field. But realistic and real-time usually can not be achieved at the same time. One of the most time consuming aspects in the entire modeling process is the creation of game geometry. Procedural modeling techniques have been used within computer graphics to create natural textures, simulate special effects and generate complex natural models including trees, buildings and so on. As a new modeling technique, procedual modeling technique has gained flourishing development in many fields such as dynamic modeling, ensemble modeling, quick modeling and so on, since 1980s when it began to be applied in the field of computer graphics. This dissertation discusses the principles of Fractal, IFS, dual-scale automation, L-System, shape grammar, split grammar and other well-known procedural modeling techniques. The characteristics, classification, development trend, suitable fields, advantages and disadvantages of procedural modeling are also discussed in detail in this dissertation. According to the research statusof applied fields of procedural modeling techniques, we combined the techniques of computer graphics, pattern recognition, image processing and physical method with the technique of procedural modeling. By exploiting to the fully their favourable conditions and avoid unfavurable theirs, we give a full play to the advantages of procedural modeling technique and the expected simulation effect is achieved. Based on the research of the core techniques and the theoretical methods, the main contributions of this dissertation are summarized as follows:(1) Researchers have been studying the simulation methods of natural scenery for a long time, which include the simulation methods of dynamic process of natural objects and the simulation methods of the shape of a natural object in virtual environment. There are many instances of dynamic simulation methods where they focused on the posture of small narrow leaves in wind. The disadvantage of proposed dynamic plant methods is that animations of vegetation and foliage are limited to simple textures in non real-time. And they can not simulate vividly the pliability of plants either. Through studying of procedural modeling methods of simulating movement of broad leaves in wind, and by combining procedural modeling methods with physical features effectively, we proposed a physically-based modeling method of plant form.The disadvantage of partical modeling is that it is difficult to deal with complicated model of tree and the plant deformation by force. To solve these problems as well as to increase the stability of simulation effect, our strategy is to take the stems, branches and leaves as a spring constraint system of multiple connecting rods to model with L-System. First of all, the segments of cylinders and cuboids are the basic units by force, which will save a lot of calculated vertexes. Then according to the Newton Motion Law, the loading procedure of stems, branches and leaves is simplified. We also set up a constraint relation according to the deforming feature of plants and the procedure of force analysis and calculation of plants is also simplified. By using the large time step of Verlet implicit integration method, we can save a lot of computing time. Moreover, the transfer force is introduced to keep the segment methods in accordance with the partical method, and it can generate more realistic movements of the plant. It is the force that the surrounding segments or particals will be affected when any part of them acted by wind. The experimentation indicates that our simulation algorithm has high performance and the shapes of the plant in wind are well comparing with the former methods.(2) The shape simulation methods of natural scenery also has been attracting the eyes of researchers, especially for the virtual plant.Virtual plant is a virtual model of plant which is built by computer based on the analysis of plants of the real world. The modeling of virtual plant includes the research of structure mechanism, growth process and the environment interaction of plants. Because the applied fields are different, the functions and implementations of software which includes procedural modeling technique are also different. The computer model used to build plant shape can be divided into two main classes. The first one is morphology generating model which generates computer shapes being approximate to the real plant in a visual perspective. This model focuses on the visual verisimilitude. By analyzing the geometric features of the structure of plants, it can draw beautiful shapes of many kinds of plants without knowing much knowledge of the botany. Another class of model is based on the knowledge of physiological ecology of plants. It main studies subjects are crops. According to the changing law when plants growing as well as a lot of experiment data, it imitates the natural growing process of plants. Though the software which can simulate the shape of plants is becoming more and more mature, software that can be used in the field of agriculture is still in the phase of studying.In orer to combine the external morphology with the physiological ecology of plants and to simplify the modeling process, we proposed a new procedural modeling method which is based on Dynamic Bayesian Network (DBN). It utilizes the powerful ability of describing uncertain prediction problems of DBN, and infers and generates an ecological network model of virtual plant to control its morphogenetic model by component interfaces. The morphogenetic model by combining parametric L-system and timed L-system guarantee the continuous growth property of virtual plant. Through comparing with the experiment data of Jichun's wheat height of Intelligent Agriculture Information Technology Application Demonstration Project, the reliability of simulation value of physiological ecology turns to be high. In order to reduce the complexity of parameter operation, storage space and computing time, we proposed a concept of Component Box L-System (CBL-System). The experimental results indicate that this approach is both effective and realistic.(3) Recently the study of architecture modeling by researchers is becoming more and more attractive. With the development of computer technology, more and more new techniques have been implemented in building construction. There are three kinds of common computer techniques involved: Digital Input Method (DIM), IBR (IBM) technology and 3D Modeling technology. The cost of the hardware (especially the scanners) restricts the broad implementation of DIM. The efficiency of IBR is not satisfied. 3D Modeling Technology includes manual modeling and automated modeling. Although manual modeling generates the most accurate and complicated models, it is comparatively time-consuming and professional knowledge in modeling field is required for the users. Then many researchers focus on the automated modeling method, but slow modeling and unautomatic modeling are the shortcomings of these methods.So we proposed a hybrid modeling method of ancient architectural of China base on local feature. To implement this modeling system, we have built a feature component model library of CAP by L system in advance. The Singular Value Decoposition (SVD) is inadequate in the detail of description and efficient identification information. Feature points have been detected by feature extraction algorithm in order to solve the problem of SVD. After obtaining the local feature regions with strong component recognition ability, these regions are presented to the Multi-scale SVD linear discriminant analysis (MSVD) algorithm which can retain both the good characteristics of SVD and the overall and partial information of images. Feature component vectors are classified by PSO-Classifier based on MSVD. Experiments have proved that this method has not only a good ability to identify categories, but also benefited the modeling speed and automation of modeling.In addition, we also introduced an assembly grammar to generate generation rules automatically. We integrated split grammar and stochastic L-system into the assembly grammar, which ensures the selection of feature components and the control of model assembly. Experiments show that the stability of the efficiency of the system can simulate realistic virtual display of the ancient architecture.(4) For the geometric models in the form of digital images transmitted through the network, we have studied the watermarking method of digital images. At present, digital watermarking methods have some shortcomings as follows: 1) the stability of the image feature point extracted is unstable, which will reduce the resistance to the attacks of random cropping attack of digital watermarking. 2) They can not use the features of image content to decide the embedded size of watermarking adaptively, which will reduce the ability of resistance to scaling attack. 3) They do not deal with the watermarking effectively and safely, which will reduce the security level of watermarking. To deal with the disadvantages described above, we proposed a method which is based on multiple scales space theory and intrinsic geometric regularity of image. This method can resist general geometric attacks including global affine transformation, loss compression, and noise and so on. It is strong robust second generation of blind watermarking algorithm. The realization process is based on Harris-Affine method to extract the feature points of image, and then use these points to build local feature regions which can be resistant to translation, rotating, scaling, and affine transformation. And we can treat the local feature regions as repeated embedded regions of watermarking. Then we can use approximation error criterion to find out the best geometric flow direction. Then we can trace the Bandelet base to inhibit the affect of noise. Last, according to the characteristics of orthogonal vectors, we embed the blind watermarking into the local feature regions to increase the security level of the whole system. We experimented under 17 kinds of attacks. The result shows that this method can resist the general signal processing attacks and geometric attack which includes rotating, scaling, cropping and linear transformation attacks. Under the image processing attacks or geometric attacks or the combination of the two, the detection side can detect the feature points and the geometric flow direction robustly. This method has advantages especially in compression and de-noising. It can protect effectively the geometric model which is being transmitted.As discussed above, the studies of this dissertation have enriched procedural modeling technique and the applied fields. They also have certain theoretical meaning and application value. We also provided effective means for the 3D computer modeling technique and its security.