近日,我协会成员撰写的论文《Recognizing irrelevant faces in short-form videos based on feature fusion and active learning》在CCF-C级期刊(Neurocomputing)上发表。第一作者为朱明成(2018级网安专业本科生),合作作者为张容川(2017级网安专业本科生),指导老师为王海舟副教授。
作者提出了一种基于深度学习特征融合算法和主动学习算法的无关人脸识别系统。首先,利用人脸识别技术将采集的视频中的人脸提取出来,并通过四种有关人识别法则标记人脸。其次,将标记到的人脸输入特征提取模块,分别提取包括人脸图像锐度、人脸尺寸在内的统计特征和使用深度学习网络Inception ResNet提取到的语义特征。随后,将两种特征相结合,共同输入融合与识别模块,进行特征融合和无关人脸识别。最后,通过主动学习模块,使多个识别模型在新的数据上进行自监督学习,从而提高整个无关人脸识别系统泛化性能。


期刊简介:
Neurocomputing是中科院2区、CCF-C级期刊;主要研究的主题涉及神经网络计算理论、神经网络计算的实践和应用。
官方介绍:Neurocomputing publishes articles describing recent fundamental contributions in the field of neurocomputing. Neurocomputing theory, practice and applications are the essential topics being covered.
Neurocomputing welcomes theoretical contributions aimed at winning further understanding of neural networks and learning systems, including, but not restricted to, architectures, learning methods, analysis of network dynamics, theories of learning, self-organization, biological neural network modelling, sensorimotor transformations and interdisciplinary topics with artificial intelligence, artificial life, cognitive science, computational learning theory, fuzzy logic, genetic algorithms, information theory, machine learning, neurobiology and pattern recognition.论文链接:https://www.sciencedirect.com/science/article/abs/pii/S0925231222008013