Computer Vision Assessment Of Fabric Wrinkle Grade

A new approach based on the technology of computer vision is proposed in this thesis to evaluate fabric wrinkle grade objectively. The content of the thesis covers three parts, i.e. 3-D reconstruction of wrinkled fabric surface shape, extraction of feature parameters from the reconstructed fabric surface shape and evaluation of the wrinkle grade of the fabric using pattern recognition, and consists of eight chapters. The brief introductions of each chapter are as follows:The foreword relates to the background of choosing the topic of this thesis.Chapter 1 gives major the summarization of overseas present situation of the research work in the field of fabric wrinkle grade assessment. The aspects involved are including the methods of measuring the surface shape of wrinkled fabric, such as instrument method, initiative stereo measurement method and surface shape recovering method using 3-D reconstruction, the extraction and selection of the feature parameters of wrinkled fabric surface and the algorithm of pattern recognition used for wrinkle grade assessment.Chapter 2 makes a description of two kinds of algorithm for 3-D reconstructing the surface shape of wrinkled fabric, one is photometric stereo method the other is shape from shading method. The basic principle and realization processing of these algorithms are all related in chapter 2. The reliability of the algorithms are tested and validated by using synthetic images, objects with simple surface shape and AATCC standard wrinkle pattern. Photometric stereo method is based on multiple illuminated images to realize the 3-D reconstruction, and is a convenient and feasible one, if the illumination conditions can be controlled. The shape from shading method can be effectively applied to reconstruct 3-D surface shape from single illuminated image only, but needs to add certain restrict condition to the algorithm.Chapter 3 recounts the extraction and selection of the feature parameters reflecting the wrinkle c haracteristics o f fabric surface. Ten different p arameters areextracted from the reconstructed 3-D surface shape of AATCC wrinkle pattern. They are contrast, power spectral density, fractal dimension, surface areas, coarse degree,twist degree, kurtosis, average offset, whole wrinkle density and sharpness. The calculation results show that the correlation coefficients between these feature parameters and wrinkle grade are all over 90 percent, which means all the parameters can be further used to evaluate the wrinkle grade by pattern recognition.Chapter 4 provides an equipment designed by author-self for capturing the illuminated images of the fabric. The images from 26 different fabric samples have been taken by using this equipment. Using these images the surface shapes of wrinkled fabric are reconstructed with the algorithms mentioned in chapter 2 and the feature parameters of these fabric samples are extracted. Meanwhile the wrinkle grades of these samples are also subjectively evaluated by three skilled assessor and the results are given in chapter 4 as well.Chapter 5 involves the evaluation of wrinkle grade using Kohonen self-organize neural network and Heb learning rules. Kohonen self- organize neural network is based on clustering principle and trained by competition. Heb learning rules are the improved algorithms of self-organize neural network for fact converge of the computation. The feature parameters of AATCC standard wrinkle pattern are input to the neural network for training and learning at first. The output neural cells are the excellent one developed after competition. Their classified number is corresponding to the wrinkle grade of the AATCC pattern, thus can be used for assessing the fabric wrinkle grade directly. Twenty-six fabric samples are assessed with the same procedure as AATCC pattern and the results are compared with those of subjective assessment given in chapter 4. It is shown that the correlation coefficient between objective and subjective assessment are 93.8% and 92.26%, and the percentage of the samples...