Font Size: a A A

Research On3D Embossed Characters Segmentation Method Based On Variable Illumination Direction

Posted on:2014-01-19Degree:DoctorType:Dissertation
Country:ChinaCandidate:W ChenFull Text:PDF
GTID:1228330398959628Subject:Mechanical and electrical engineering
Abstract/Summary:PDF Full Text Request
In the process of industrial applications, there is a great variety of identification marks on different materials to be detected.The work pieces, representing surface deformations, are required in many production processes as durable markings.These embossed characters is recorded not by colour difference, but through the different reflection of light between character region and background region.Because of their three-dimensional structure, characters created this way are often difficult to illuminate, to segment and, consequently, to detect. With the development of automatic vision inspection, it has put forward higher requirements for the accuracy of the character recognition.The segmentation of embossed characters from the background image plays an important role in the recognition system. In the study of how to make use of the3D feature of embossed characters to achieve segmentation, some overseas scholars have presented some new segmentation methods, while the domestic research is still concentrated in the segmentation based on two-dimensional gray-scale image.Three different segmentation methods is proposed in this paper according to the three-dimensional feature of character segmentation, based on variable illumination direction.The main research is as follows:Firstly, in the Illumination space, by clarity comparison of image acquired in diffuse illumination and directional illumination, directional illumination is more appropriate to inspect3D embossed characters. However, this kind of3D embossed characters presents a different appearance under different illumination directions. So the depth information is inadequate with2D gray image in single direction acquisition for3D object surface. The scheme is proposed that extract surface normal vector feature using the relationship between two-dimensional surface gray image in variable illumination direction for subsequent segmentation.After optimization of the direction of the light source, the light azimuth angle distribution (0°,90°,180°and270°) and the optimum elevation angle (63.43degrees) is determined.The image acquisition system is designed with lighting, image acquisition device and the four direction of light source synchronous lighting acquisition circuit.The background lighting difference is removed using polynomial regression planes, and gray level uniform between images is eliminated with histogram equalization.It improves the precision of surface vector calculation,laid a foundation for subsequent embossed characters segmentation.Then, three different embossed stereo character segmentation methods are proposed based on four variable illumination direction systems.In the segmentation method based on surface texture, the concept of texture features based on the surface normal vector is proposed.The fuzzy set theory is introduced to LBP operator, and Fuzzy-SNLBP operator segmentation method is proposed based on texture features.The influence of various parameters in Fuzzy-SNLBP to the segmentation results is analyzed. According to the comparison experiment:Fuzzy-SNLBP has more information than the SNLBP, fuzzy can effectively reduce the SNLBP calculation noise; Fuzzy-SNLBP is better in texture resolution than LBP method using the gray value and SNLBP with fixed threshold.The segmentation method based on surface normal vector and the graph clustering is proposed.The graph is created based on surface normal and graph clustering segmentation algorithm is was executed.In the segmentation process, the feedback control theory is introduced to optimize image segmentation coefficient.The2D entropy of binary image as the segmentation quality evaluation criterion,the segmentation coefficient is adjust by the feedback control.The optimal segmentation result (segmentation coefficient k=64, when2D information entropy S=2.0681is the minimum value);A simple segmentation method without illumination consistency processing for the original image is proposed based on pulse coupled neural network(PCNN) and image fusion.After extract the highlight of two-dimensional gray image acquired in variable illumination direction, through the image fusion method, get the segmentation of the complete binary image, its framework for subsequent recognition. The segmentation evaluation criterion for embossed stereo character are builded in this research, and the parameters of PCNN model are dynamic adjusted.Through the experiment determined the optimal segmentation parameters.Finally, using the receiver operating characteristic curve (ROC) and the character recognition rate, each of the three segmentation algorithm for character segmentation results is evaluated. The evaluation results show that:three segmentation algorithm of3D embossed characters based on four variable illumination direction have more advantages than single light source.Their recognition rate is higher than other methods of image segmentation based on single light source.The character recognition rate based on graph theory clustering segmentation method is the highest, but the execution time is also the longest; identification method based on PCNN was the lowest, but the implementation of the shortest time; texture segmentation method execution time and recognition rate is centered on.It can be selected according to the need in practical application.
Keywords/Search Tags:Embossed stereo character, Image segmentation, Variableillumination direction, Surface normal, Fuzzy-SNLBP operator, Graph clustering, Pulse coupled neural network, Segmentation evaluation criterion
PDF Full Text Request
Related items