利用特征距离信息引导决策融合的多模态生物特征识别方法

传统的决策层融合作为识别系统最末端的融合层次,具有信息量不足的缺点,对于各模态分类性能差异较大的系统,识别率低且可靠性差。提出了一种基于特征距离信息的决策层融合方法,应用于包含虹膜、手掌静脉和手指静脉的多模态生物特征识别系统。以置信度作为权重,通过权重来探索不同模态生物特征识别的性能差异,实现了有效特征信息的提取,并且提高了系统的抗干扰能力。该方法充分考虑了权重因子与特征距离信息和模态分类性能参数之间的复杂关系,将模态的决策偏好通过置信度转化为定量表征,不仅使各模态权重因子的求解更具科学性,而且提高了识别系统在复杂情境下的自适应能力。实验结果表明,该融合方法的识别精度与抗干扰能力优于其他决策层融合算法。

Multimodal Biometric Recognition Method based on Decision-Level Fusion guided by Feature Distance Information

As the last fusion hierarchy in the recognition system, traditional decision-level fusion has the disadvantage of insufficient information. Especially for those systems differed in classification performance, the recognition effect is not very ideal. In this paper, a novel confidence factor based on feature information was proposed , and iris, palm and finger vein were used for fusing at the decision level in multimodal biometric identification system. In this article, weighting factor was used to explore the difference of performance between modalities, to implement the effective feature detection and improve the anti-interference ability of the system. This method fully considers the complex relationship between the weights and modal classification ability and the score information. Though confidence factor, it transforms the decision preferences into the quantitative characterization. Not only it makes the identification results more effectively in complex situation decision, but also it makes the solution of the model weight more scientific and objective. The experimental results show that the proposed fusion method is superior to other fusion algorithms.