A Fine Grained Digital Textile Printing System And Key Techniques

Fabrics include non-planar fabrics and printed fabrics.Printed fabrics are different from non-planar fabrics for their rich colors,various patterns and low costs.Non-planar fabrics are considered more attractive,expensive in appearance and luxurious owing to solid texture patterns,floats and luster contrasts.Typical non-planar fabrics include embroidery,corduroy,jacquard and so on.To produce fabrics which are both colorful and lavish,and have color gradients,fine grained digital textile printing is developed.Fine grained digital textile printing is to print digital patterns onto non-planar fabrics to produce graceful and colorful fabrics.It is a novel technology which combines the digital textile printing and non-planar fabrics.In this thesis,we focus on this technology to produce colorful and lavish fabrics with solid texture patterns.The key problem with textile printing is the distortion caused by directly printing ink-design pattern onto non-planar fabric gives poor performance.To address the problem of dis-tortion,we develop a solution to accurately align ink-design pattern with fabric.Our approach consists of three procedures:1)taking pictures of fabric,2)employing image registration based computer vision technology to deform ink-design pattern according to deformability of the fab-ric,3)printing deformed ink-design pattern onto the fabric.By incorporating the emerging research efforts into the computer vision,we target at new techniques supporting fine printing on three different fabric types,namely,embroidery with stitches inside shape,embroidery with only stitches along shape,and corduroy.The approaches proposed in this thesis have been im-plemented and verified by our prototypes.In summary,the contributions of the thesis are as follows:(1)We present a novel fine-grained digital printing system based on image registration tech-nology that produces digital ink-designs onto non-planar fabrics.This solution consists of specific hardware and suitable algorithms to overcome the deformability of fabric.There are three key hardware components in the proposed printing system,including an indus-trial camera,an LED,a computer and a digital textile printer.The industrial camera is used to capture high-resolution image of fabrics.The LED is to ensure same illumination condition for all images.The computer is to implement fabric image segmentation,reg-istration of binary images of fabric and digital pattern and deformation of digital pattern.(2)We propose a fast and accurate image registration algorithm based on key points corre-spondences for embroidery with stitches inside shape.Firstly,we get binary images of fabric and digital ink-design pattern by texture image segmentation.Secondly,multi-scale weighted patch matching is employed to register two binary images.Finally,we estimate warping parameters according to matching result and deform digital ink-design pattern.(3)We present an image registration algorithm for embroidery with only stitches along shape.By image segmentation we get binary images of fabric and digital ink-design pattern,ex-tract local shape context descriptors for key points and employ improved Jaccard simi-larity coefficient to match key points.Then we use L1-median to get precise matching result.For skipped stitches for fabric,we use locally linear embedding(LLE)to calculate the positions of the missing key points.(4)For corduroy with stripe texture,we present an image registration algorithm.The key technique is the image segmentation method which is based on frequency-domain filter-ing for images with stripe texture and image smoothing.The stripe texture on corduroy is easily mistaken for foreground pattern.The approach of image segmentation proposed in this thesis consists of three phases.Given an image,the effect of stripe texture is weakened by filtering in the frequency domain automatically.Then,structure-preserving image smoothing is employed to remove texture details and extract the main image struc-tures.Last,we use an effective threshold method to produce segmentation results.