A Research For Geometry Shape Representation For Profiled Fibers

Fiber recognition has made an important role in textile importing, exporting, manufacturing and selling. But the quality is unsteady and the processing is time-consuming when the inspection is carried out manually. In order to improve the quality and efficiency of fiber inspection, a research of automatic fiber inspection is held by Shanghai Entry-Exit Inspection and Quarantine Bureau and Donghua University. By means of computer image processing and pattern recognition, fiber inspection and form classification are processedautomatically.With the geometry shape representation, profiled fibers are classified by means of their cross-sectional shapes. This provides essential. information for the study of profiled fibers' characteristics. Profiled fiber image description is applied to extract shapes information and remove noise and interference that is induced in cross-section preparation and image acquisition. Fiber shape representation data should be packed separately for every fiber since the processing thereafter is repeated for each fiber. The novel shape description of multiscale triangle area representation (TAR) locates concave and convex segments by calculating the area which is determined by three sample points in object boundary. In the proposed algorithm, TAR calculation is implemented and normalized at various scales. TAR Average Filtering (TAR-AVG) and TAR Median Filtering (TAR-MED) are discussed to deal with concave and convex property calculation. Although TAR-AVG is more precise, the application is limited. TAR-MED behaviors better than TAR-AVG, but many useless outputs contaminate the resultCombining with the two methods, a new method called TAR Median Average Mixed Filtering (TAR-MAMF) is proposed. The calculation of TAR is modified by revising the sinusoid, and the ameliorated method is called TAR-MAMFW. The character parameters are obtained from histogram outputs, which are counted by the data in each column.Relying on those parameters, the geometry shapes of profiled fibers arerecognized.In experimental results, shapes from normal profiled fiber database as well as overlapping profiled fiber database were computed by TARAVG, TAR-MED, TAR-MAMF and TAR-MAMFW. Then, those four methods were also applied in MPEG7_CE-Shape-1_Part_B database on different categories of shapes. The results demonstrate that the TAR-MAMFW is accuracy and can find applications in various fields...