研究目的
To develop a robust hyperspectral face recognition method by combining local features extracted via low-rank tensor decomposition and global features via polar discrete fast Fourier transform, and to create an ensemble classifier for improved accuracy.
研究成果
The proposed method effectively combines local and global features for hyperspectral face recognition, achieving competitive accuracy compared to existing methods. It highlights the complementary nature of spectral and spatial information, but suggests future work to enhance discriminative information and ensemble techniques.
研究不足
The study is limited by the small public hyperspectral face databases available, potential intra-person variations over time, and computational complexity due to high data dimensionality. The local ensemble classifier did not show significant advantage over the global classifier, indicating room for improvement in patch selection and ensemble methods.
1：Experimental Design and Method Selection:
The study uses a novel ensemble classifier combining local features from tensor decomposition and global features from PFFT. It involves patch selection using Fisher ratio, tensor decomposition for dimensionality reduction, and PFFT for contour feature extraction.
2：Sample Selection and Data Sources:
The Hong Kong Polytechnic University Hyperspectral Face Database (PolyU-HSFD) is used, with 25 subjects, 33 spectral bands from 400-720 nm, and images of 220x180x33 pixels.
3：List of Experimental Equipment and Materials:
Hyperspectral camera with CRI's VariSpec liquid crystal tunable filter (LCTF) and halogen light system.
4：Experimental Procedures and Operational Workflow:
Images are preprocessed, patches are selected, local features are extracted using tensor decomposition, global features via PFFT, and an ensemble classifier is built using particle swarm optimization (PSO).
5：Data Analysis Methods:
Recognition rates are calculated using ten-fold cross-validation, and comparisons are made with methods like PCA, SVM, CRC_RLS, and PLS.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
