Research Of Fabric Weave Pattern Based On Empirical Mode Decomposition

Recognition and reassignment for weave pattern is akey issue in the field of computer aidedfabric analysis and design. Accurate recognition of weave pattern has constituted the scientificbasis for textile imitative design and innovativedesign, which will significantlypromotecompetitiveness of textile industry.Traditional methods, which are characterized by geometric, spectral or model methods, areall based on the stationaryassumption for the fabric image. This means that the weft and warpyarnsare arranged in a regular spatial way with the certain gap between the yarns in warp-orweft-directions. However real weave patterns are mostly spatially-variant, i.e. the fabric imagesare non-stationary. Considering the disadvantages of traditional methods, an adaptive schemeusing empirical mode decomposition (EMD)for the recognition of fabric weave pattern isproposed, which is followed by fabric weave pattern reassignment algorithms. The proposedscheme first iteratively decomposes the underlying fabric image into a number of intrinsic modefunctions (IMFS). The first order IMF is applied to histogram computation and mean shiftalgorithm and fabric weave pattern segmentation results are obtained.As segmentation results of fabric weave pattern, pattern grids are applied to construct newwave patterns. The combination of traditional image deformational algorithm with spatial linearinterpolationtechniqueor virtual amplification technique yields fabric weave pattern deformationresults which are spherical deformation, rippling deformation etc. The main contribution aspectsare as follows:(1) On the study of fabric pattern segmentation, an adaptive method based on bidimensionalempirical mode decomposition is proposed. The first intrinsic mode function, which mainlyimplies the weave pattern of the fabric,is applied to construct the histogram. Togetherwithcorresponding threshold decision strategies, the segmentation is performed effectively.(2) An effective segmentation algorithm is presented by extending classical mean algorithmto EMD domain.The proposed method first iteratively decomposes the underlying fabric imageinto a number of intrinsic mode functions (IMFs). The first order IMF is used to operate themean shift algorithm. Simulation results show that EMD based mean shift method is a promisingapproach for the segmentation of fabric weave pattern.(3) Certain methods regarding the fabric texture distortion and synthesis have beenintroduced in this paper too, including thedistortion,water wave deformation, synthesizing and mirroring, which aim to obtain abundant fabric patterns. Simulation results for fabric texturedistortion and synthesis are given to demonstrate the functionality of the proposed approaches.