Yarn Evenness Detection Methods Based On Image

The detection of yarn filament irregularity is one of main indices to measure yarn quality. At present, there are mainly two methods to detect yarn irregularity: instruments measurement method and visual measurement method. These methods each have advantages and disadvantages. With the development of computer image processing technique, taking full advantage of digital image processing technique to overcome disadvantages of other detect methods and improving the level of yarn quality detection has become an important research field and has important significance to enhance yarn quality.This paper mainly studies image preprocessing in yarn evenness detection. We further propose two methods that process the yam blackboard images, based on the limitation of existing research.In this method, yarn images are obtained by scanning blackboard which is twisted by the yarn. Our intention is that an image with minimal distortion is got and target signal is completely detached from background signal, i.e. wiping off the noise and standing out the target. At present, Mathematical Morphology is the mainly way to process this kind of images, but there is no effective method target against the yarn images with more hairiness. Variational partial differential equation is an important image denoising method. Total variation model is the representative method of PDE. The greatest advantage is that it can preserve boundary as well as denoising. Recent years, curvelet transform has attracted related researchers' attention, especially in the field of image processing. Curvelet, as a new multiscale analysis algorithm, is more appropriate for the analysis of the image edges such as curve and line characteristic than wavelet, and it has better approximation precision and sparsity description. The advantages of the two image processing methods were adopted to propose two methods. One is based on the combination of Mathematical Morphology and image decomposition based on variational PDE, and the other is based on curvelet transform. In order to detect accurately, this paper correct the skew the scanning images automatically, avoiding the calculation error caused by image tilting. Experiments show that the new methods make the best of the image structure and texture, yielding denoised images with better visual quality especially against yarn images with more hairiness. Furthermore, we use it to detect yarn evenness, which is coincided with that of visual measurement method.