基于梯度方向直方图特征的掌纹识别关键技术的研究

发布时间：2018-11-26 07:37

【摘要】：随着智能化、信息化与社会、生活各方面的不断融合、交互,互联网和物联网的日益普及,信息和系统的安全性问题成为备受瞩目的关键性问题,身份认证作为解决安全性问题重要手段之一,受到广泛的关注。生物识别技术便应用而生,掌纹识别具有特征丰富稳定具有可靠性和唯一性、用户易接受、易获取等优势,近几年已成为人机交互和模式识别等领域的重点研究对象。传统的掌纹特征提取和识别技术在识别精度和速度方面仍存在着许多不足,特征提取和匹配至今为止仍是学者们研究的重点,需要对其进行进一步的改进和性能的提升。本文通过阅读大量掌纹识别相关文献,了解国内外研究及发展现状,归纳总结、对比分析传统算法的优劣势,针对特征提取和模式匹配问题,采用梯度方向直方图特征(HOG),并结合分区的分块二值模式和压缩感知算法将其用于掌纹识别方法。本论文的主要工作如下:(1)提出基于分区的分块二值模式与梯度方向直方图特征的掌纹识别方法。该方法主要采用的是纹理特征和边缘特征的融合特征,充分利用二者互补的特性来提升算法的性能。首先,对原掌纹图片进行预处理获取掌纹感兴趣区域。然后,分别提取掌纹R0I区域的分区MB-LBP特征和HOG特征。将得到的分区MB-LBP特征和HOG特征串联起来,得到图片融合后的特征。最后,使用最近邻分类器对图片进行分类,得到识别结果。使用北京交通大学的掌纹库进行实验后,通过与传统算法进行对比分析,本文的算法在识别精度上具有相对的优势。(2)提出基于压缩感知与梯度方向直方图特征的掌纹识别方法。首先,对原掌纹图片进行预处理得到掌纹感兴趣区域,提取R0I区域的HOG特征,将训练样本的HOG特征作为稀疏表示的过完备字典。然后通过COMP算法求解图像在过完备字典上的稀疏表示,求得一组最优稀疏系数重构每一个图像,最后计算测试样本图像HOG特征矩阵与各类重构图像的最小残差得出分类结果。使用北京交通大学的掌纹库进行实验,表明本文算法不局限于小样本的情况,具有较高的识别性能。
[Abstract]:With the continuous integration and interaction of intelligence, information and society, various aspects of life, and the growing popularity of the Internet of things and the Internet of things, the security of information and systems has become a key issue that has attracted much attention. As one of the most important methods to solve the security problem, identity authentication has been paid more and more attention. The technology of biometrics has been applied, palmprint recognition has the advantages of rich, stable, reliability and uniqueness, easy to accept by users and easy to obtain. In recent years, palmprint recognition has become an important research object in the fields of human-computer interaction and pattern recognition. Traditional palmprint feature extraction and recognition techniques still have many shortcomings in recognition accuracy and speed. Feature extraction and matching are still the focus of scholars' research so far, it needs further improvement and performance improvement. In this paper, we read a large number of palmprint recognition related documents, understand the domestic and foreign research and development status, summarize, compare and analyze the advantages and disadvantages of the traditional algorithm, aiming at the problem of feature extraction and pattern matching, we adopt gradient direction histogram feature (HOG),. Combined with partitioned binary pattern and compression sensing algorithm, it is used in palmprint recognition. The main work of this thesis is as follows: (1) A method of palmprint recognition based on partitioned binary pattern and gradient direction histogram feature is proposed. The method mainly adopts the fusion features of texture feature and edge feature, and makes full use of the complementary characteristics of the two features to improve the performance of the algorithm. Firstly, the original palmprint image is preprocessed to obtain the region of palmprint interest. Then, the partition MB-LBP features and HOG features of palmprint R0I region are extracted. The segmented MB-LBP features and the HOG features are concatenated to obtain the features after image fusion. Finally, the nearest neighbor classifier is used to classify the images and the recognition results are obtained. The palmprint database of Beijing Jiaotong University is used to carry out the experiment, and by comparing with the traditional algorithm, The algorithm in this paper has relative advantages in recognition accuracy. (2) A palmprint recognition method based on compression perception and gradient direction histogram features is proposed. Firstly, the region of palmprint interest is obtained by preprocessing the original palmprint image, and the HOG feature of R0I region is extracted. The HOG feature of the training sample is regarded as an overcomplete dictionary with sparse representation. Then, the sparse representation of images in overcomplete dictionaries is solved by COMP algorithm, and a set of optimal sparse coefficients are obtained to reconstruct each image. Finally, the classification results are obtained by calculating the HOG feature matrix of test sample images and the minimum residuals of various reconstructed images. By using palmprint database of Beijing Jiaotong University, it is shown that the proposed algorithm is not limited to small samples and has high recognition performance.
【学位授予单位】：内蒙古农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】