Applications of Computer Vision in Fashion and Textiles

Calvin Wong

• Publisher: Woodhead Publishing
• Year: 2017
• Print ISBN: 9780081012178
• eBook ISBN: 9780081012185
• Topic: Technology & Engineering
• Subtopic: Materials Science
• Index: Technology & Engineering

Applications of Computer Vision in Fashion and Textiles provides a systematic and comprehensive discussion of three key areas that are taking advantage of developments in computer vision technology, namely textile defect detection and quality control, fashion recognition and 3D modeling, and 2D and 3D human body modeling for improving clothing fit. It introduces the fundamentals of computer vision techniques for fashion and textile applications, also reviewing computer vision techniques for textile quality control, including chapters on wavelet transforms, Gibor filters, Fourier transforms, and neural network techniques. Final sections cover recognition, modeling, retrieval technologies and advanced human shape modeling techniques.The book is essential reading for scientists and researchers working in the field of fashion production, quality assurance, product development, textiles, fashion supply chain managers, R&D professionals and managers in the textile industry.- Explores computer vision technology with reference to improving budget, quality and schedule control in textile manufacturing- Provides a thorough understanding of the role of computer vision in developing intelligent systems for the fashion and textiles industries- Elucidates the connections between human body modeling technology and intelligent manufacturing systems

Part One

Fundamentals of common computer vision techniques for fashion and textile applications

1

Fundamentals of common computer vision techniques for textile quality control

J.L. Jiang⁎; W.K. Wong^{† ⁎} Nanjing University of Information Science and Technology, Nanjing, China
^† The Hong Kong Polytechnic University, Kowloon, Hong Kong, China

Abstract

In current apparel industry, almost all apparel-manufacturing plants still rely on human visual method of fabric inspection. However, the accuracy of human visual inspection can only achieve around 60%–75% and is limited to identifying obvious fabric defects. To improve the effectiveness of fabric inspection, automated fabric inspection system based on technology of computer vision is an emerging trend and a great demand from the textile and apparel industry. This chapter first introduces current challenges of common computer vision techniques for textile quality control. Then basic system architecture of an automated fabric inspection system and methodologies of computer vision for fabric defect detection are presented. Finally, merits of computer vision technique and its prospects for fabric defect detection are discussed.

Keywords

Fabric defect detection; System architecture; Statistical approaches; Spectral approaches; Model-based approaches

1.1 Introduction

As we know, the three basic human needs are nutrition, shelter, and clothing. This explains why the textile industry is as old as human civilization. Fabrics are closely related to human life; not only are they the key materials for clothing, but they also can be applied to many industrial products. Since the beginning of the industrial revolution in the 18th century, woven fabric production has been accomplished by sophisticated weaving machines, which operate mostly autonomous and uncoupled from any human interaction. No production process is 100% defect-free because other than the fiber quality, the price and the quality of the fabrics depend on the amount and the occurrence of defects.

Currently, upon receipt of fabrics from the textile mills, apparel manufacturers rely on fabric inspection machines operated by a trained fabric inspector. During the inspection, the fabric roll is loaded into a fabric inspection machine, and the fabric roll can be released and mounted on the inspection surface while the fabric is in moving status. Once the fabric inspector identifies a defect on the moving fabric through visual inspection, he or she stops the machine and records the defect with its location. Although human visual inspection can carry out the fabric inspection effectively, there are many drawbacks too:

(1) An operator needs to be trained for a long time to become a skilled fabric inspector.

(2) Visual inspection is a very tedious task even for a skilled fabric inspector. They inevitably miss small defects and sometimes even large ones after long working hours.

(3) Human inspection is not reliable as a fabric inspector needs to assess the fabric normally with 1.6–2 m width on the moving status with around 10 m per minute. It is extremely difficult for a fabric inspector to concentrate for a long period of time. Thus, the manual fabric inspection, which the current industry greatly relies on, has only an accuracy rate of around 60%–75%. Hence, inspection efficiency is low.

To increase the effectiveness and efficiency of fabric inspection, an automated fabric inspection is necessary to save labor costs; achieve higher inspection, accuracy, and efficiency; as well as provide real-time inspection data.

This chapter will introduce the fundamentals of common computer vision techniques for textile quality control. Current challenges of common computer vision techniques for fabric defect detection will be presented in Section 1.2. Basic system architecture of an automated fabric inspection system will be presented in Section 1.3. Methodologies of computer vision for fabric defects detection will be given in Section 1.4. Current benefits of computer vision technique and its prospects for fabric defects detection will be discussed in Section 1.5. The conclusion of this chapter will be summarized in Section 1.6.

1.2 Current challenges of common computer vision techniques for fabric defect detection

Fabric defect detection is a very complex task and research in this field is ongoing. The implementation of a common computer vision techniques-based method for fabric defect detection online may still show the following difficulties:

(1) The fabric defect detection is particularly challenging due to the existence of many kinds of fabrics. For different types of fabrics, different algorithms are required.

(2) The algorithms have to be effective and can be implemented in real-time. Unfortunately, some of them are effective for fabric defect detection but are computationally complex for online applications, while the others may be implemented in real-time but are ineffective otherwise.

(3) The characterization of defects in fabrics is generally not clearly defined.

(4) The overdetection rate is very high. Although visual inspection systems at present can achieve high detection accuracy, they are still difficult to be used in weaving industry due to the intolerable overdetection rate.

(5) The impact of noise. During image acquisition, noise will be more or less introduced, which will also affect the accuracy of fabric defect detection.

1.3 Basic system architecture of an automated fabric inspection system

The automation of a visual inspection process [1] requires complex interaction among various system components. The typical automated fabric inspection system mainly includes a set of cameras, frame grabbers, a computer, and a lighting system.

Two types of cameras are used for the fabric defect defection: line-scanning and area scanning. Line-scan cameras contain a single row of pixels to capture data. As the object moves past the camera, a complete image can be reconstructed in software line by line. A line-scan camera can expose a new image while the previous image is still transferring its data, which differs from area-scan cameras. The advantage of the line-scanning camera is that it provides high resolution and can be generally analyzed by texture-based methods. The disadvantage is that it does not generate complete image at once and requires external hardware to build up images from multiple line scans. Area-scan cameras capture an image of a given scene on account of a matrix of pixels. They are more general purpose than line-scan cameras. Compared with line-scan cameras, area-scan cameras can image a defined area quickly, whereas a line-scan camera must be moved over the area to produce a similar image. In a sense, line-scan cameras can capture clearer texture images than area-scan cameras, but the cost of a line-scan camera is very high; therefore, an array of area-scan cameras is commonly used for economical fabric defect detection.

A frame grabber is an electronic device that captures individual digital image from a digital video sequence. It is an important component of a computer vision system, in which video frames are captured in digital form and then stored, transmitted, or processed. In fabric inspection system, the data coming from all cameras are transformed into a digital image by the frame grabber simultaneously. One way to cope with multiple cameras is to use one frame grabber unit per camera, which permits parallel processing of image pixel data. Finally, the capture images are transported to the computer.

Data procurement by the frame grabbers is downloaded into the host computer, which mainly performs two functions: defect detection and classification. Defect detection will be achieved by the corresponding defect detection algorithms. After the completion of defect detection, we should score the defects and classify them according to the demands of the factories. At present, the defect classification in the textile industry mainly depends on traditional manual assessment. The cloth inspectors score the fabric defect and evaluate the fabric grade by the scores. However, this manual work is time consuming, and involves problems of subjective human factors, physical and psychological load and fatigue, and professional experience. With the rapid development of computer vision, automatic and efficient methods for fabric classification are desperately needed by which the size of defect can be measured automatically. Based on the common defects, a defect database can be established. Based on the detected defect, a pixel-matching-based algorithm is used to measure the size of the defect. The linear/nonlinear pixel-matching-based algorithm can be used for the linear/nonlinear defect. The defect can be scored with a rule based on the size of the defect and then the fabric classification can be implemented automatically.

Lighting is very important for image collection since it decides quality of the acquired images. The image quality is drastically affected by the type and level of illumination. We usually use two types of lighting for visual inspection: front lighting and back lighting. Back lighting is mainly used for avoiding shadow or glare problem.

1.4 Computer vision-based techniques for fabric defect detection

The key of computer vision-based techniques for fabric defect detection lies in the defect detection algorithm. At present, there are mainly three kinds of defect detection algorithms: statistical approach, spectral approaches, and model-based approaches. The structure of this part can be seen in Fig. 1.1.

Fig. 1.1 The different defect detection algorithms.

1.4.1 Statistical approach

Statistical approach is mainly applied for the analysis and interpretation of data to measure the spatial distribution of pixel values. In fabric defect detection, we usually assume that the statistics features of defect-free parts are stationary, and the feature of defect parts can be separated by a fixed or a dynamic threshold. In Mahajan et al. [2], classified the statistical approach into first-order, second-order, and higher-order statistics. The first-order statistics that we usually encountered are the mean and variance, which ignore the spatial interaction between image pixels. Compared with the first-order statistics, second- and higher-order statistics consider two or more pixel values relative to each other. Statistical approaches in fabric defect detection can be described by representations such as co-occurrence matrix (CM), mathematical morphology (MM), fractal method, and so on.

1.4.1.1 Co-occurrence matrix

As one of the most popular statistical texture analysis tools for fabric defect detection, the principle of CM is based on repeated occurrences of di...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Contributors
• Part One: Fundamentals of common computer vision techniques for fashion and textile applications
• Part Two: Fabric defect detection
• Part Three: Investigation and evaluation on fibers, yarns and textile quality
• Part Four: Fashion and textile recognition, modeling and retrieval
• Index