[论文解读] Extracting Region of Interest for Palm Print Authentication

Kasturika B. Ray|arXiv (Cornell University)|Dec 21, 2013

Biometric Identification and Security参考文献 9被引用 7

一句话总结

本文提出了一种用于掌纹认证的新型感兴趣区域（ROI）提取方法，以提升识别准确率。通过识别并隔离掌纹图案的中心区域——纹线、线条和皱纹，该方法提升了基于纹理的生物特征识别系统的性能，在集成ROI后实现了识别准确率的显著提升。

ABSTRACT

Biometrics authentication is an effective method for automatically recognizing individuals. The authentication consists of an enrollment phase and an identification or verification phase. In the stages of enrollment known (training) samples after the pre-processing stage are used for suitable feature extraction to generate the template database. In the verification stage, the test sample is similarly pre processed and subjected to feature extraction modules, and then it is matched with the training feature templates to decide whether it is a genuine or not. This paper presents use of a region of interest (ROI) for palm print technology. First some of the existing methods for palm print identification have been introduced. Then focus has been given on extraction of a suitable smaller region from the acquired palm print to improve the identification method accuracy. Several existing work in the topic of region extraction have been examined. Subsequently, a simple and original method has then proposed for locating the ROI that can be effectively used for palm print analysis. The ROI extracted using this new technique is suitable for different types of processing as it creates a rectangular or square area around the center of activity represented by the lines, wrinkles and ridges of the palm print. The effectiveness of the ROI approach has been tested by integrating it with a texture based identification / authentication system proposed earlier. The improvement has been shown by comparing the identification accuracy rate before and after the ROI pre-processing.

研究动机与目标

• 提升基于掌纹的生物特征认证系统的准确率。
• 从掌纹图像中识别并提取一个有意义的感兴趣区域（ROI），以捕捉最具区分性的特征。
• 开发一种简单、有效的ROI提取技术，适用于与现有基于纹理的认证系统集成。
• 通过定量比较，评估ROI预处理对识别性能的影响。

提出的方法

• 所提出的方法基于纹线、线条和皱纹的分布，定位掌纹活动的中心区域。
• 它提取一个以主模式区域为中心的矩形或方形ROI，最大限度减少无关背景噪声。
• ROI提取在完成标准化预处理步骤（如归一化和对比度增强）之后进行。
• 该方法设计为计算高效，并在不同掌纹图像上具有鲁棒性。
• 提取的ROI随后用作先前认证系统中基于纹理的特征提取模块的输入。
• 通过比较集成ROI前后识别准确率的变化来评估性能。

实验结果

研究问题

• RQ1 如何从掌纹图像中自动提取一个有意义的感兴趣区域，以提升认证准确率？
• RQ2 ROI预处理对基于纹理的掌纹识别系统性能有何影响？
• RQ3 一种简单、非复杂的ROI提取方法是否能有效提升多样化掌纹样本的识别准确率？
• RQ4 该ROI提取方法对掌纹方向和图像质量的变化是否具有鲁棒性？

主要发现

• 与使用完整掌纹图像相比，所提出的ROI提取方法显著提升了识别准确率。
• 基于ROI的方法减少了噪声和无关特征，从而实现了更可靠的特征提取。
• 该方法在不同掌纹样本中均有效，并保持了ROI位置的一致性。
• 将ROI预处理步骤与现有基于纹理的系统集成后，识别性能实现了可测量且一致的提升。