The Study On Multi-Focus Textile Fiber Images Fusion

Textile fibers have different strength, cross-sectional width, length and curvature. Diversity also exists in the same type of fibers. These characteristics are important criterions to determine maturity of fiber, so they are critical data during preprocessing. Thus studying the image of textile fibers has great significance for accessing and analyzing of these characteristics information. There are many fibers in the same scene, these fibers are scattered in different depths of view. This caused multi-focus problem in textile fiber images. There are different focal effect in different parts of the fiber texture images in the same scene. The multi-focus problem in fiber images interferes the result of image processing and further interferes the accuracy of fiber parameter calculation. In this thesis, image fusion technique is applied to attack this problem.These is redundancy and complementary information in a series of images of the same scene with different focuses, it is possible to obtain clear sub-regions from source images that are under the same imaging conditions and reconstruct into a result with better focus. More abundant information and more accurate characterization of targets would be obtained in the result image than original. Thereby, it facilitates the next processing, evaluating and decision-making of fiber targets.Image fusion methods are commonly classified into spatial domain methods and transform domain ones. Spatial domain methods include the logic filter, morphology, binary image algebra, etc. Transform domain methods include pyramid algorithm and wavelet transform and so on. These existing techniques have their advantages. However, several drawbacks had been found. There would be unstable results brought by too many necessary parameters; Blurring effect, misregistration and missing information might be caused by improper threshold selection; low accuracy result would emerge due to sensitivity about noise; Non-real-time performance would lead to low-efficiencyIn order to construct a more clear and comprehensive image of textile fibers in real-time, a novel image fusion method should be proposed. Through this method which based on the features of fibers, the characteristics of the fiber objects can be further identified and researched. Among source images in the same scene, there is redundancy and complementary information of time or space. This technology use this information to get a image which have clear focus in individual fibers.Thus, this thesis proposed a image fusion method based on local clarity. With the model of pixel gray value, it could measure the level of pixel clarity. First of all, by searching maximum modulus values of multi-focal images, the layer number of the maximum clarity pixel (that is, the maximum model of gray value) would be got and stored in a matrix. Second, to deal with the disturbance of noise, the proposed scheme uses the threshold depended on the maximum modulus values in local region and amends the matrix of layer number. Then, according to layer number matrix, the pixel gray values of corresponding layers synthesized to get the result image. Finally, some measures used to improve processing speed. To evaluate and prove the performance of the proposed method, series of experiments and comparisons are carried out in this paper. Experimental results on real textile fiber images indicate that the proposed method is effective.