Researching And Realizing On The Towel Label Defect Detection Algorithm Based On The OpenCV Technique

With the leap development of industrial technology,people pay much library,causing it plays a vital role in the fields of computer vision.For nearly 10 years,OpenCV’s continuously updates has made it become more and more functional and powerful due to its portability which combines with the all kinds of hardwares like DSP or ARM micro-chips that contains digital processors or chips with high quality.OpenCV is now widely used in mainly fields like industrial automation field,military field,the field of remote sensing satellite and transportation,as in this paper,which applies it to the product surface quality detection field.As similar to the vast majority of image processing applications,this topic research which focused on the label surface quality detection including the main process,such as the image pre-processing,image feature extraction and its matching(which is used for the image registration for this paper),and finally,extracting the potential or existed surface defects by using the corresponding morphological methods or operations.In this article,the pre-processing procedures mainly includes the image filtering operation,histogram statistics of image,threshold segmentation and binarization or morphological processing methods.Realizing the image registration is the critical step during the process of defect detection.The registration accuracy whether is high or not affects the subsequent defect extraction operation directly.The main work of this paper is as follows:Firstly,introducing the origin of the "machine substitutes for human "development strategy and its development trends.Combining with this paper’s research based on the history from the past till the current stage of towel label included in the printed-product defects detection.Secondly, this paper expounds the development course of OpenCV technology,modules and its important applications on image processing.Mainly includes the simple narration and overviews.Thirdly, pointing out the shortcomings of the classical Scale Invariant Feature Transform (SIFT)algorithm to the image feature extraction and matching on the basis of the image feature extraction framework.Then came up with an updated version of SIFT algorithm Speeded Up Robust Feature algorithm combined with the template matching method on the surface defect detection.Comparing SIFT with SURF algorithm’s difference of extracting feature keypoints.This paper used the two methods to extract the standarded-label’s local invariant feature keypoints and the defected-label’s,which is invariant to the image rotation,translation,scaling and even the perspective change.Next step is using the Random Sample Consensus Algorithm to purify the characteristic points and obtaining the corresponding coordinate information of the points,estimating the image registration parameters according to these information.After that,making the relative transformation with the parameters to realize the image registr-ation between the standarded-label image and the defected-label(to be detected) image.Fourthly,extracting the defects after acquiring the registered image.This paper proposed the two methods.One is based on image grayscale difference diagram,the other is based on XOR binarized image’s pixel value.The experiments proves the latter method could drop some defect information that will interfere with the final defect decision,so the former is more optional.Finally,summing up the research results of this paper and looking forwarding to the future based on the deficiency of this research so that it can be a guidance to the future work.