A Comprehensive Review of Dimensionality Reduction Techniques and their Performance Comparison in Image Processing Applications

Abstract

The rapid growth of image processing and artificial intelligence applications has resulted in the generation of extremely high-dimensional datasets. Images obtained from medical imaging systems, satellite sensors, surveillance systems, and industrial inspection devices contain a large number of features, many of which are redundant, irrelevant, or noisy. High-dimensional data increases computational complexity, memory consumption, training time, and the risk of overfitting in machine learning and deep learning models. To address these issues, dimensionality reduction techniques are widely used as preprocessing methods to reduce the number of input features while preserving meaningful information. This review paper presents an in-depth analysis of dimensionality reduction techniques used in image processing applications. Both feature selection and feature extraction methods are discussed, including Correlation-Based Feature Selection, Forward Feature Selection, Linear Discriminant Analysis (LDA), Principal Component Analysis (PCA), and Empirical Mode Decomposition (EMD). Furthermore, modern deep learning-based dimensionality reduction approaches such as Autoencoders [8] and Convolutional Neural Networks (CNNs) are reviewed. A comparative performance evaluation is provided based on classification accuracy, computational efficiency, and robustness to noise, scalability, and suitability for various image processing tasks. The paper also highlights the role of dimensionality reduction in medical image diagnosis, face recognition, remote sensing, and object detection systems.