Automatic Speckle Region Detection For Digital Speckle Pattern

Digital image correlation(DIC)is a non-contact full-field deformation measurement method,which has the advantages of convenient operation and low requirements for test environment.It has been widely used in aerospace,civil engineering,biology and some other fields.In recent years,requirements for high-precision and real-time measurement are gradually increasing,and the amount of experimental data is increasing.This is not only a high demand for measurement hardware,but also a new challenge for the software algorithm of image data processing.How to effectively reduce the data redundancy of one single image and improve the automatic performance of DIC has become one of the problems that cannot be ignored in the research field of DIC.This dissertation presents a solution to this problem from the point of view of automatic selection of speckle regions.The basis of speckle region selection is to analyze visual components of the DIC image.Analysis shows that appropriate features and anti-interference mechanism should be incorporated in each specific solution.This dissertation presents the relationship equation between speckle region selection and speckle assessment criteria,and establishes the framework of speckle region selection.This dissertation starts with the features of digital speckle patterns,establishes a set of evaluation system of speckle area detection,and analyzes the advantages and disadvantages between common features in image processing and features based on speckle assessment criteria.By comparing these parameters,it is found that modulus sum of the local intensity gradient vector,the local entropy,and the local mean floating are recommended.At the same time,the feature parameters with the calculation template of 13 pixels × 13 pixels to 21 pixels × 21 pixels can be optimized considering accuracy and efficiency.Based on the features of speckle patterns,this dissertation introduces the variational level set method into speckle region selection for the first time,which overcomes the shortcomings of traditional methods.The discrete iterative scheme of local Gaussian distribution(LGD)method is derived.Combined with the conclusion of the second chapter,the automatic LGD method is proposed to realize the automatic selection of the initial region.In order to further improve the robustness of this algorithm,morphological filter is added to fix “small holes” in the image caused by uneven speckle preparation.The test results on DIC images show that the automatic LGD method greatly improves the calculation accuracy and stability of region detection,and the average error at the boundary is less than 1 pixel theoretically.The automatic LGD method has been integrated into an open source DIC software to achieve the effect of automatic deformation calculation.Although the calculation accuracy and stability of the automatic LGD method have been improved,calculation time of this method is rather undesirable.The initial region fast grabbing method greatly reduces the amount of redundant calculation,and the Gauss filter is used to further improve the stability of the calculation.The improved automatic LGD method not only ensures the calculation accuracy,but also greatly improves the calculation efficiency.In this dissertation,the feasibility of parallel computing strategy in speckle region detection is verified,and block computing parameters are proposed.The research results show that the automatic LGD method can achieve the detection speed of 3-5 frames/s while ensuring the localization accuracy.