项目名称: 部分遮挡下人脸人耳融合识别方法研究
项目编号: No.61300075
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 袁立
作者单位: 北京科技大学
项目金额: 22万元
中文摘要: 从身份识别的角度来说,人脸与人耳具有明显的互补性,这使得人耳信息可以成为人脸识别的有益补充。但在实际应用,遮挡是目前困扰其识别性能的一个主要因素。由于稀疏表示理论符合人类在感知图像时的稀疏性机制,对噪声和遮挡具有较好的鲁棒性,所以本项目拟将该理论与生物特征识别应用相结合,以人脸人耳融合识别为基础,利用遮挡在图像上的分布具有的稀疏特性,从理论上研究如何消除部分遮挡带来的影响。本项目将重点研究:(1)如何利用已有图像数据来学习得到对应于遮挡和无遮挡情况的超完备字典?(2)如何建立稀疏表示特征提取框架与模型,使得所提取的特征向量同时具有鉴别能力和稀疏性,并能够根据实际应用实现稀疏性的自适应调控?(3)如何改进稀疏表示分类器的鉴别能力较弱和稀疏求解计算量大的问题?(4)如何实现鲁棒的的人脸人耳融合识别?
中文关键词: 人脸人耳融合识别;部分遮挡;稀疏表示;;
英文摘要: For persoanl identification, obviously complementary physical position of face and ear make the human ear a very helpful supplement to facial recognition. In real application, eliminating the influence of occlusion turns out to be a more demanding work. Sparse representation shows robustness to noise and occlusion in image coginiton. So in this project, we prosose robust non-intrusive face and multimodal recognition based on sparse representation (SR). The main contents of this research project include: (1)how to design the SR over-complete dictionary for occluded images and non-occluded images on the basis of the given training samples? (2)how to construct the framework and model of feature extraction method based on sparse representation? In this model, the feature vectors extracted will be discriminative and sparse. The sparsity of the feature vector is adjustable. (3)how to improve the performance of sparse reprentation classifier to posses higher disciminative ability and lower intensive computation? (4) how to design robust multimodal recognition based on face and ear
英文关键词: multimodal recognition using face and ear;partial occlusion;sparse representation;;