Objective Assessment Of Fabric Pilling Grade Based On Wavelet Transform

Fabric pilling will seriously affect their appearance, style and feel, and reducesthe performance of fabric, therefore the fabric pilling resistance is an importantindicator of textile detect. The current assessment methods are mostly subjectivemethods with a lot of subjectivity, evaluation results will be affected by personalfactors, and can not precise or quantify describe the fabric pilling degree. This articlewill use the digital image processing technology, according to the newer methodssuch as wavelet analysis theory, fabric texture energy algorithm and neural networkmethod to process pilling fabric images. And exploring the the methods to objectiveassessment and quantify describe fabric pilling performance by using digital imageprocessing technique. Promoting the assessment of fabric pilling grade developtowards an objective direction.In the process of objective assessing fabric pilling grade by using digital imageprocessing technology, the fabric texture filtering, pilling segmentation and extractionof the pilling characteristic value is the foundation of the pilling rating. In this paper,we are mainly to analysis and research the fabric texture filtering, pillingsegmentation, the pilling characteristic value extraction and neural networks rating.Firstly, we have a preliminary pre-processing to samples, according to thecharacteristics of pilling fabric image obtained by digital cameras. Stretching the grayhistogram of the samples to eliminate the light uneven, and eliminating image noise,provided foundation for the subsequent processing.In the process of fabric texture filtering, we combining the wavelet analysistheory and fabric texture energy algorithm, calculating the fabric texture energygradient of fabric images after multi-scale wavelet decomposition, and fitting thecurve of the energy gradient, automatically obtain the wavelet reconstruction level,according to the energy gradient curve, then accurate reconstructing the pilling imageby wavelet reconstruction, and obtain the pilling image after texture filter.In the process of pilling segmentation, the paper analyzed and compared thesegmentation principle and segmentation results of edge detection operators,mathematical morphology segmentation algorithm and threshold segmentationalgorithm. Finally the method of combining iterative threshold segmentationalgorithms and mathematical morphology algorithm is chose for our pillingsegmentation, and split out complete pilling binary image. At the same time, thepilling binary image feature extraction is proceed, then normalized the 11characteristic parameters extracted and analyzed the pilling level correlation,according to correlation coefficient, we selected the characteristic parameters ofpilling number, total pilling area, pilling density, the average height of pilling, totalpilling volume, the average volume of pilling and the roughness as indicators ofpilling grade assessment.Finally, this article researched the neural network identification model used in the objective assessment of pilling fabric grade, input the characteristic parametersselected to BP neural network for training, and input actual pilling fabric samples fortesting. The ultimate proof of the objective assessment of the fabric pilling ratingmethod used in this article is feasible.