A Study On Multi-Focus Nonwoven Web Image Fusion Technology Based On Fiber Natural Boundray

The performance of nonwoven web is influenced by fiber diameter, porosity and orientation obviously. Traditional detection techniques are time-consuming, costly while the detection using computer image technology tends to be fast and can ensure accuracy. The computer image technology relies highly on the clarity of fibrous material image. However, due to the small depth of field, light microscope can ensure the clearness of the target partly while other parts fuzzy, resulting multi-focal phenomenon of the ultra-depth of objects (i.e. thin nonwovens, yarn and uneven fiber sections). Therefore, it is important to solve the problem "how to integrate multi-focal images into a new clear image", so that light microscope images focused at different positions can be fused into a clear completely image for the subsequent processing.Multi-focal image fusion based on the pixel level is to fuse a clear image with a set of images, extracting complementary information and eliminating redundant information. It retains the information of original scene, provides rich and clear information of the target, guarantee the reliability of the source for the subsequent processing. Now pixel-level multi-focus image fusion technology is divided into three algorithms:the multi-resolution algorithm, the single-pixel-based algorithm and the region-based algorithm. Image fusion based on single pixel is simple, fast, but ignores the relation between pixels. The multi-resolution image fusion is complex, processing slowly and its decomposition layers depend on the image resolution despite showing the excellent effect. Moreover, it is easy to bring ghosting and a large number of error information in the merged image. The algorithm based on region is the method with two steps:divides the image into several local blocks, and uses some parameters to choose the reasonable layers to replace the region correspondingly. The method may make the focus and blurred both in one region if the block is too big, while causing the zigzagging in case of the small block.An optical microscope image acquisition system with automatic focusing function was used in this paper to capture a series of microscope images on different focal planes. The system controls three motors on XYZ axis to achieve the movement of the microscope platform. The sample can be focused at different focus position by the z-axis’s movement. Different DOF and the num of layers were set according to different materials (i.e. nonwoven web, fiber and yarn).As to fuse textile fiber materials more reasonable, this paper analyzes the image fusion algorithms domestic and abroad, proposed a region-based image fusion algorithm with multi-source-points’diffusing based on fiber natural boundary:firstly, the source points were selected and the diffusion of the source points was began under the limitation of the fibers’ boundaries. Secondly, the diffusion stopped when the diffusing edge of a source point couples another; the diffusion stopped when the diffusing edge reached the fibers’ boundaries. The region division was determined till the diffusing finished. Finally, each region was flood with the best focused corresponding region from the series of microscope images and the final fused image was formed. The fused image was verified by evaluation methods and experiments.1) In order to obtain fiber natural boundary, this paper firstly deals with multi-focal images in pre-processing including gray, smooth image denoising, etc. Then different methods of the sharpness’ calculation and holes filling are compared to select the more excellent one. Finally, fibers’ natural boundaries are reconstructed by point-based image fusion, image segmentation and flood.2) The rules of the source-points’ selection were studied in this paper. Considering the different characteristics of the background and objectives, a fast and adaptive threshold was determined to select the source points, decreasing the difficulty of the subsequent processing.3) Within the fibers’ boundaries, the source points were diffusing in8-neighborhood. The diffusing stopped when the diffusing edge of a source point reached the fibers’ boundaries; the diffusing also stopped when the corresponding edges were constructed by the different source points. The image was divided into different regions when the diffusion was finished.4) The appropriate regions from the series of the images were selected to construct the fused image. Four parameters were chosen to select the layer for the region as the principle.5) According to four fusion evaluation criteria, this paper evaluates the effect of fusion images based on different algorithms. And the different samples were analyzed. The result indicates that the method of image segmentation in this paper is a verified effective way to reduce the large plaque effect and increase the target information. Meanwhile, a series of experiments on pore size distribution and orientation of nonwovens were taken to verify the clarity and the accuracy of the fusion image.