The Research On Recognition Of Complicated Fiber Image

China is the biggest textile producing and export country. In the past, the identification for the classification of fiber mainly relied on manual labor, which led to identification error owing to the complicated operation steps, low efficiency and influence of human factors. So the automatic recognition for the fiber by the computer is becoming tendency.In this thesis the recognition of complicated fiber image is focused. As the fibers in the image are intertwined and covered each other, the existing methods are difficult to achieve the recognition. The proposed recognition method includes three parts: edge detection, edge thinning and repair and the match of edge curves.The multiscale edge detection algorithm based on wavelet transform is first presented, which can detect edge for the preprocessed fiber image. The maximum points of magnitude in different scale are calculated by wavelet transform. These points are corresponding to the mutation points. The possible edges are obtained by using the maximum points of magnitude. The pseudo edges caught by noise are removed by the adaptive threshold. The edges under different scale are fused and the edges of fiber are obtained.The edge curves by edge detection have certain width. The edge curves must be thinned to the single pixel width for the purpose of feature match. The combination of Zhang thinning algorithm and the lookup table is proposed for the thinning of fiber edges. As the cover of other fibers led to the break of fibers and the thinning algorithm leads to the small burr on the some nodes, the thinned edge curves are post processed, which can achieve the full fiber edge curves.After obtaining the fiber edge curves, the edge curves are matched each other, which can recognize the complete fibers. The coarse to fine curves matching algorithm is proposed. The coarse match is achieved by using the distance between the feature points of two curves. The least-squares method is used to fitting curves. The fine match is completed by the compare of the coefficients of quadratic polynomial.The experiments show that the proposed method can recognize the complex fiber image. The recognition rate is about 80% on the sample images.