Font Size: a A A

Research On Key Technologies And Algorithms For Palmprint Recognition

Posted on:2009-10-15Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y X WangFull Text:PDF
GTID:1118360272484605Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the development of digital network and communications technologies and the constantly expanding of physical and virtual space of the human being, the demand on the safety for system and information is growing. Under the informational, networked and digital conditions, traditional personal identification methods having taken on some main faults which are not insurmountable, so they can not meet the demand for the development of society. Under the case, biometrics, which refers to automatic identification of a person based on his/her distinct physiological or behavioral characteristics, was emerged. Biometrics makes biotechnology and information technology to merge into each other. Biological characteristics possess many desirable features. For example, they are innate for everyone, different individuals have different characteristics and they will remain unchanged for a long time. Moreover, Biological characteristics can not be forgotten or lost and possess innate convenience and efficiency in technology. As an important tool for personal identification and verification, the interest in biometrics has increased dramatically.As an important part of the biometrics, palmprints exploit the effective information on our palm for automated personal authentication. Palmprints possess unique virtues and rich and stable features: palmprints contain more distinctive information; main features of palmprints are stable and distinct; palmprints are difficultly faked up; palmprint capture devices are much cheaper; the capture method of palmprints is easy; palmprints have a high level of user acceptance. It is for these reasons that palmprint recognition is becoming a hotspot in the biometric field.This dissertation concentrates on the research of the key technologies and algorithms in palmprint recognition field. Firstly, a set of palmprint image capture device is developed. Then, according to the characteristics of palmprints, we propose a series of efficient algorithms: palmprint preprocessing methods, feature extraction and matching algorithms. Finally all of algorithms are used to construct a palmprint identification system. The main creative work in the dissertation is:1. Propose a robust and effective palmprint alignment and segmentation algorithm. Because the place of palms and the degree of stretching of fingers are not limited, the difficulty of location and segmentation of palmprint images is added. To resolve theses problems in palmprint preprocessing, a definition of the bitwise switch frequency is given in the algorithm, then a local effective area is determined by the statistic value of the bitwise switch frequency of the binary image of a palmprint image, and, in the area, the edges of fingers are completely separable, so three key points can be located by tracking the boundary. Next, using one side of the boundary of a palm as the reference line aligns the palmprint. Then, the potential area which includes the center of the largest inscribed circle is determined by the coordinates of these key points in the aligned palmprint image, which needs less time to search the largest inscribed circle of the palm and increases the speed of the algorithm. Lastly, a coordinate system of the palmprint is constructed and a ROI area is extracted from the central part of the palmprint image. The proposed palmprint alignment and segmentation method can effectively avoid the invalidation of the boundary-tracking algorithm, improve the accuracy of alignment, decrease the influence of the position of fingers on the accuracy of alignment, adjust the size of the ROI area by the size of the different palms automatically and extract more effective information from the palmprint image.2. Considering the characteristics of palmprint images, a steerable filters based fuzzy unsharp masking algorithm is presented to enhance the contrast of a palmprint image. Because palm lines are a kind of roof edges, and they are irregular and have different directions, the proposed method chooses unsharp masking technology as main body, uses steerable filters to replace the Laplacian filter to extract high frequency information of palmprint image along different directions. In addition, palm lines also have different depths in a same palm, so the fuzzy set theory is introduced into the unsharp masking method and a half open fuzzy membership function is designed to distinguish high frequency and medium frequency and low frequency components in the palmprint and control their degree of enhancement. In the method, the enhancement of noises in the flat area is restrained by reducing the degree of enhancement of the low frequency components; the contrast of wrinkles is effectively enhanced by increasing the degree of enhancement of the medium frequency components and the degrees of enhancement of principal lines and wrinkles are balanced by weakening the degree of enhancement of high frequency components. The proposed method not only has the capability to induce resistance to noises, but also can enhance the contrast of principle lines and wrinkles synchronously and can eliminate the "shadow effect" in the palmprint image. 3. Propose an extraction method of palmprint algebraic features using kernel Fisher discriminant analysis. The methods of using subspace analysis for palmprint recognition have becoming the mainstream methods. These methods can obtain better recognition rate, when there are enough training samples. But when there only are fewer samples, the recognition rates obtained by these methods decline. In addition, some nonlinear factors are introduced in the course of capturing palmprint images, such as the shift, rotation and distortion of palms. Thus a satisfied accuracy can not be obtained by existed linear subspace analysis based palmprint recognition methods. To resolve these problems, Kernel Fisher discriminant analysis method is firstly applied in palmprint recognition and Kernel Fisherpalms based palmprint recognition method is proposed. In the method, a palmprint image is mapped into high-dimensional feature space F by utilizing a nonlinear map. The algebraic features of palmprints called Kernel Fisherpalms are extracted and used to classify the palmprints in the feature space. Because of considering more high-order statistical information of palmprint images in the feature space, the Kernel Fisherpalms based palmprint recognition method can more effectively distinguish different palmprints. Firstly, the parameters in the method are determined by experiments. By using the method with these parameters, a satisfying accuracy can be obtained.4. Proposes a line structure features extraction and matching method. Firstly, considering the features of palmprint images, global filtering stage and local filtering stage are designed for extracting palm lines with different directions and depths in the dissertation. In the different stages, steerable filters with different numbers and directions and different threshold values are applied for different objects to extract continuous and whole palm lines. On the base of the above works, a method of line structure features representation and matching for palmprint recognition by using improved water-filling algorithm is proposed. The method firstly extracts cross point features from palmprint images explicitly and combines them with other extracted global structure features and local structure features to achieve the representation and matching of palmprints. The extracted structure feature of palm lines not only can represent global structure, but also can provide direction information of the points in the palm lines and the distribution of different type cross points in different local regions. These features include the position information of palm lines, global structural information and local structural information. Thus, using the extracted structure features can effectively distinguish different palmprints with similar structure.5. Propose a Dual-Tree Complex Wavelet Transform based Local Binary Pattern Weighted Histogram (DT-CWT based LBPWH) method for the extraction of palmprint texture features. The method combines two effective texture analysis tools to enhance the discriminant capabilities of extracted textrue features from different palmprints. In addition, using the local directional characteristics and significance of different sub-regions of palmprints to compute a weight set to improve the discriminative ability of the final histogram vector. A training procedure is unnecessary to construct palmprint model in our approach, that is, the method is entirely independent on the training set. The experimental results demonstrate that the proposed approach not only can give a better performance, but also can meet the need of the palmprint recognition system for real time. On the basis of these above works, the preprocessing methods and DT-CWT based LBPWH method are integrated into a palmprint recognition system. The application system can implement the capture of palmprint images and personal identification. The palmprint recognition system developed by the dissertation, no matter consider from the use cost the system, or in terms of the degree of complicacy of the system, can well satisfy the actual applied demand.
Keywords/Search Tags:Personal Identification, Biometrics, Palmprint recognition, Feature Extraction, Feature matching
PDF Full Text Request
Related items