A Robust Hybrid Feature Extraction Technique for Enhancing Image Classification Accuracy

Abstract

Image classification is a fundamental task in computer vision with wide-ranging applications including medical diagnosis, remote sensing, surveillance, and multimedia retrieval. The performance of image classification systems largely depends on the quality of extracted features, which must be discriminative, robust to noise, and computationally efficient. Traditional feature extraction methods often struggle to capture both global structural information and fine-grained local patterns simultaneously. To address this limitation, this paper proposes a robust hybrid feature extraction technique that integrates complementary feature descriptors to enhance image classification accuracy. The proposed approach combines texture-based, shape-based, and statistical features into a unified feature representation. Local Binary Patterns (LBP) are employed to capture local texture information, Histogram of Oriented Gradients (HOG) is used to represent shape and edge structures, and statistical color moments are utilized to describe global image characteristics. Feature normalization and dimensionality reduction are applied to improve efficiency and reduce redundancy. The resulting hybrid feature vector is evaluated using standard classifiers, including Support Vector Machines (SVM) and k-Nearest Neighbors (k-NN). Experimental results on benchmark image datasets demonstrate that the proposed hybrid feature extraction technique significantly outperforms individual feature-based methods in terms of classification accuracy, robustness, and generalization capability. The findings confirm that combining complementary features provides a more comprehensive image representation, leading to improved classification performance.