Study And Realization Of Intelligent Fashion Design System

Intelligent Fashion design system is designed for non-professionals to do the design work traditionally done only by the profession designers. There is still no commercial system for intelligent fashion design. This thesis discusses the system architecture of the system, the main intelligent technology of the system including the abstraction of the fashion part information, discovery of fashion style knowledge and the automatic generation of the fashion style. The realization of the system is given at last. All in all, the main important research is as following:The characteristic of art itself leads to the difficulty of its knowledge expression and many possibility of system design. In this thesis we have studied three different system architecture: Fashion Prototype Based Design (FPBD), Fashion Parts Parameterization Oriented Design (FPPOD) and Mental Image Oriented Design (MIOD). As they can neither fully express the fashion design knowledge nor provide good support to the design process, while agent has the characteristics of different knowledge encapsulation, distributed intelligence, self controlled etc., we suggest agent based system design and regarded it as the optimum on the basis of three other ways' contrast. It takes fully consideration of design process, the designer's experience, user demands and the knowledge expression and operation of fashion storage and change; it is the basic guarantee of the final realization of our intelligent fashion design system.The economic storage of the fashion style's outline is studied in the thesis. For the problem of fashion information saving a new B-spline fitting algorithm under given error based on Genetic Algorithm was provided. It uses special mutation operator as the way of data point deletion to reach the goal of less storage for a given fashion style under a certain error given. As the Least Square Method(LSM) used for B-spline fitting can only make sure the least MSE for the corresponding data point not for the whole data point set and also the efficiency of the algorithm is quite low due to the too many times of fitting, two improved algorithm was provided for better precisionand higher efficiency.As we want to generate the fashion according to certain style, the style definition for each fashion turns to be important. We propose a DNA coding based algorithm for fashion style knowledge discovery. Since most of the fashion style decision variables are continuous, the discretization algorithm of the variables of the decision table is also discussed in the thesis. We use an elegant style decision table for the validity confirmation of the algorithm.This thesis proposes a way to solve the problem of automatic fashion styles design. The particularity of fashion styles design leads to difficulty of knowledge expression and various probability of technology, A GA search method coding based on the experts' knowledge is given in this thesis, which uses artificial neural network to approach fitness function. It is proved to be useful, the automatic learning and evolution of the system can be done by users' interaction.We study and propose the formalization of the fashion style for intelligent fashion design. Some of the realization details of the system are given at the end of the thesis. The main contributions of this thesis include:1. A novel agent based system architecture for major parts based intelligent fashion design is proposed and has been implemented.2. The economic storage of the fashion style's outline is well solved. A new B-spline fitting algorithm under given error based on Genetic Algorithm is proposed. Two improved algorithm is proposed for better precision and higher efficiency. Case study shows that it works well.3. A DNA coding based algorithm for fashion style knowledge discovery is proposed. It will fully support style based intelligent fashion classification both theoretically and technically.4. We have studied the fitness function's expression in automatic fashion generation based on GA. We...