Sparse bidirectional two-dimensional principal component analysis for feature extraction

The use of computer vision technology to achieve recognition has become an important research topic in the field of artificial intelligence, and feature extraction is an important step in automatic recognition using computer vision technology. The spatio-temporal feature descriptor extracts a feature vector with excessively high dimensionality, which increases the computational complexity of various classification tasks and extends the processing time. PCA aims to map high dimensional data to low-dimensional space through linear transformation and is widely used in the fields of dimensionality reduction and feature extraction, data compression, and pattern recognition. However, PCA has shortcomings, such as losing the inherent two-dimensional features of the image and producing principal components that are unfavorable for interpretation. To address these issues, this paper proposes a sparse bidirectional two-dimensional principal component analysis algorithm (S2D2PCA). This algorithm is developed by applying sparse constraints to the principal components of the bidirectional two-dimensional principal component analysis algorithm (2D2PCA). The S2D2PCA algorithm extracts features directly from two-dimensional images and preserves their two-dimensional features; improves the interpretability of the principal components, reduces redundancy, and improves the efficiency of feature extraction. In this paper, experimental validation is conducted on the ORL and Weld datasets. The results demonstrate that the classification performance of the S2D2PCA algorithm is superior to that of PCA and 2D2PCA.