Research On 3D Ear Reconstruction And Recognition Based On Laser Triangulation Measurement

Biometrics authentication which combines the information technology and biology is utilized in personal identification based on physiological or behavioral characteristics.Human ear with characteristics of universal,unique and permanent,is an ideal candidate for personal identification.Ear recognition can be widely used in national security,finance and insurance,social welfare,electronic commerce and other related fields.With the advances in the three-dimensional imaging technology,the focus of researchers is shifting from the traditional two-dimensional ear recognition to three-dimensional ear recognition.Compared to the 2D ear image,3D ear model has richer spatial geometry information.This advantage can enhance the security of the system.In addition,new 3D imaging techniques may reduce the sensitivity of the scale and illumination influence which may impact the 2D imaging quality,so that the robustness of the system can be improved.As one of the most reliable Biometrics technologies,3D ear recognition has attracted extensive attention from researchers for its research value and application prospect.Although the existing research work on 3D ear biometrics has achieved fruitful resu lts,but the there are still some unresolved issues: First,most of the public 3D ears' databases are collected by using commercial 3D scanners with an expensive price and bulky size.Besides,their software development kits are always unavailable.All these are bottlenecks for practical applications of 3D ear biometric system.Secondly,the existing methods mainly focused on the internal features extraction and recognition of 3D ear,while they ignoring that the external features can improve both the efficiency and accuracy of 3D ear biometrics.Finally,most of the research work on 3D ear biometrics are focused on a single feature extraction and recognition.Multi-feature extraction and fusion are not yet researched adequately.Considering all above unsolved issues,we are going to research some key issues of a complete system for 3D ear modeling and recognition,including the 3D ear reconstruction system,preprocessing study,feature extraction and recognition,fusion,et al.This thesis is focused on the following issues:Firstly,based on the analysis of the existing 3D imaging technologies and the requirements of the 3D ear data collection,a 3D ear reconstruction system is designed and developed according to the laser triangulation measurement.The system overcomes the disadvantages such as expensive,bulky and inconvenient for system integration of the commercial laser scanners.The system is designed for ear specially.It presents a high accuracy and high speed as well as low cost.By optimizing the imaging model,the system parameters,the system components,and the super-resolution of points cloud.The 3D ear model can be reconstructed effectively.A 3D ear database with 2000 high quality samples from 500 individuals was established.It's the biggest among all existing 3D ear databases reported.Secondly,the preprocessing,3D ear segmentation and normalization issues have been resolved.The preprocessing methods were studied according to the characteristic of original images.Laser lines were extracted using the weighted centroid method with a morphological generated mask.The K-Sparse clusting and morphological processing methods are studied for automatic segmentation of the ear region.A novel ear detection method was performed based on the relative l ocation of the endpoints on laser lines.The region of interest(ROI)can be located for feature extraction and recognition.Meanwhile,the projection density has been defined to normalize the 3D ear.Thirdly,a novel ear-parotic angle feature was proposed.According to the observation and analysis,the ear-parotic angle feature is defined and extracted respectively.Both the least squares method and the principal component analysis are used for extracting the normal vectors of the spacial points set.After the simulation and comparison,a fast and effective normal vector extraction method based on PCA is proposed.Experimental results show that,as an external feature,the angle feature is stable and helpful for improving the efficiency and accuracy in the indexing recognition of 3D ears.Finally,muti-feature extraction and fusion issues in 3D ear recognition are solved.The class of global features(center shape,angle)and the class of local features(point,line,area)were defined comprehensively.Effec tive recognition results(EER=2.2% on 2000 samples)were achieved by features optimizing and fusion.In this thesis,an automated 3D ear recognition system has been achieved,including 3D ear acquisition and reconstruction,system optimization,automated segmentation of 3D ear,3D feature extraction and recognition.Experimental results prove the effectiveness of above methods.All these studies laid a good foundation for 3D ear recognition not only in research but also in application.