| Author: | Qu, Hong |
| Title: | Intelligent systems for digital pattern analysis and design support |
| Advisors: | Mok, P. Y. (SFT) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Dressmaking -- Pattern design Tailoring -- Pattern design Fashion design -- Computer aided design Hong Kong Polytechnic University -- Dissertations |
| Department: | School of Fashion and Textiles |
| Pages: | xx, 207 pages : color illustrations |
| Language: | English |
| Abstract: | Nowadays, consumers are empowered, trendy, and they are always looking for newness and excitement. To address this need, the demand for new products is growing at accelerated speed, and this can be applied to all types of consumer products, including fashion. Graphics and patterns play an essential role in fashion design and product development. Hundreds of decorative patterns are required to be produced within days while the tedious, manual and laborious creation process is not catching up well with the demand. Very few research has been done on elevating the decorative pattern design process and the existing Computer-Aid Design (CAD) systems are rarely used to create patterns in digital format. In the digital era, computers are now being used to generate designs, but the designs or artworks generated neither meet the aesthetic criteria nor are in the correct format for production or further editing. Following the development pathway of artificial intelligence, the way we can build an intelligent pattern design system is by enabling computers to learn from design samples, of which useful pattern knowledge can be analyzed, extracted and processed to create new designs. By thoroughly reviewing the existing practices of design creation in the fashion industry, a new framework is proposed. Three intelligent systems are developed and experimentally verified in this study. They include the efficient repeated pattern detection system (R-system), the automatic design element extraction and vectorization system (E-system), and the vector-based digital pattern generation system (G-system). Each system serves a specific purpose. They can work either interpedently as standalone system or seamlessly work together as a whole for digital pattern creations. The R-system is responsible for efficient detection of repeated patterns from input images. A repeated pattern is the smallest unit that can tile the overall image and represents the primary information of the image in a compact form. Repeated patterns appear in many kinds of visuals: from natural scenes, building and architectures, to designs such as textiles, product/graphic designs for packing or wallpapers. Detection of such repeated patterns also supports many downstream applications such as image retrieval and image synthesis. After reviewing different approaches, a hybrid method is proposed to keep a good balance among content diversity, topology regularity and the trade-off between robustness and speed. In particular, this study firstly leverages activations of a pre-trained Convolutional Neural Network (CNN) to predict coarse repeated pattern size options of the input image. Accurate repeated patterns of the input images are then obtained by template matching optimization. Experiments are conducted on a proposed dataset to demonstrate the superiority of our methods. The accuracy of our method is 0.673 which is 20% higher than the baseline method and the time cost is only its 11%. The proposed R-system analyzes input images to detect repeated patterns generating unit region of image with repeated patterns or entire image of unrepeated image as outputs. Such outputs from the R-system can be inputted to the E-system for further design analysis. The E-system focuses on automatically identifying and extracting core design elements from input images. A design image can be considered as an arrangement of design elements filled with specific colors according to certain layout rules. This study vectorizes the extracted design elements and keeps outcomes compact. In particular, unsupervised segmentation is applied during core design element extraction to solve the problem of the lack of dataset for labeled designs. Next, a novel design element deconstruction method is proposed for vectorization based on color quantization. Extensive experiments on design images demonstrated the effectiveness of the proposed method. Our method extracts and vectorizes the core design elements of an image in around 13 seconds. The output vectorized design elements are more compact than common business software and are easier to reuse for new design generations. Furthermore, a vector-based design element dataset is built to support design generation. The proposed R-system and E-system can analyze large volume of digital pattern samples, and extract useful design elements and rules such as colors and topology structures. With the learnt design rules, the G-system is proposed to support the creation of new vector-based patterns, meeting humans' need on aesthetics and are ready for production. The proposed G-system is for generation of new digital patterns that can be applied in textiles and other decorative graphic designs. To achieve this, this study generally classifies textile patterns into three main groups: stripe, check, and motif. New designs can be regarded as combinations of different geometric and color parameters. Hence parametric models are formulated by flexible combination of geometric and color parameters learnt from reference images. Implementation experiments prove the design outputs meet basic human aesthetic, and our system can support design work. To summarize, a new REG framework is proposed in this study by integrating three intelligent systems for analyzing and supporting pattern design making. It promotes the use of artificial intelligence for design generation that meets the needs of human aesthetic requirements and caters to the speedy fashion product development cycles. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12404