|dc.description.abstract||The textile industry needs a reliable, accurate, affordable, and efficient system to automatically evaluate fabric wrinkles. This thesis is a preliminary attempt to achieve this goal.
In this work, one laser-camera system is set up to capture images of fabrics with wrinkles. These images are called range/depth images since their gray values are proportional to heights and depths of fabric wrinkles. Two algorithms are designed, namely the facet model algorithm and the plane-cutting algorithm, to extract features for evaluating wrinkles on fabrics and on six replicas, which are the standard plastic specimens representing different levels of wrinkles.
In the facet model algorithm, the input image is the original range image captured by the laser-camera system. This input image is first processed by a notch filter. Then it is passed through two low-pass filters with different window sizes, generating two blurred images. A difference image is obtained by subtracting the two blurred images. After that, the facet model is used and the difference image is treated as a piecewise-continuous gray level surface around each pixel by applying bivariate-cubic polynomials to approximate this surface. Then topographic categories on the continuous surfaces are calculated. The pixel areas of hillside slopes, ridges, peaks, ravines and pits are summed up as a feature to describe the grades of wrinkles.
Based on the facet model algorithm, the plane-cutting algorithm is developed. A difference image is obtained in the same way as in the facet model algorithm. The difference image is dissected by a series of horizontal planes at different range levels. These planes are called cutting-planes. Then the intersected areas between the image and cutting-planes are exploited to be a new feature to evaluate wrinkles.
Through setting proper parameters such as the blurring window sizes and the facet model window sizes, both algorithms successfully classify the six replicas. But for the fabric specimens, the wrinkle grades from both algorithms are not consistent with those assigned by technicians. The effects of hairy fabric surfaces interfere in evaluating wrinkles of fabric specimens. Finding better solutions to evaluate fabric wrinkles requires further research.||