Painting style classification and art image analysis model by fusion of convolutional neural network and principal component analysis

Abstract

To address the challenges of poor adaptability to multi-scale features and high-dimensional redundancy in painting style classification, this study proposes an improved classification method by integrating the Xception network with Principal Component Analysis (PCA) for dimensionality reduction. The Xception model extracts both fine-grained details and global semantic information, while PCA selects the most discriminative principal components to reduce redundant features, enhancing classification accuracy and efficiency. Experiments conducted on Eastern and Western painting datasets with 10-fold cross-validation demonstrate that the proposed model outperforms state-of-the-art methods, including EfficientNet, RegNet, and ConvNeXT. Notably, it achieves an average classification accuracy of 0.973 in Western painting classification. Moreover, the integration of PCA significantly reduces computation time, with classification speeds of 105 ms for Eastern paintings and 100 ms for Western paintings, surpassing benchmark models. This study presents an efficient and precise solution for automated art style classification, demonstrating the effectiveness of combining deep learning with dimensionality reduction. The findings offer valuable insights for computational art analysis, supporting applications in artwork identification, digital archiving, and cultural heritage preservation.