A Novel Technique for Still Image Classification Using Hybrid Learning: Benchmark Dataset

This article presents a novel hybrid learning approach that combines hand-crafted features, including Scale-Invariant Feature Transform (SIFT) and Haralick texture, with deep-activated VGG19 features for image classification. The extracted features are used to classify objects into multiple categories through various machine learning algorithms, such as Decision Tree, Naive Bayes, Random Forest, and XGB classifier. Experiments on the Caltech-101 dataset, which contains noisy, rotated, and rescaled images, show the proposed method achieving an outstanding 99.13% recognition accuracy using the Random Forest classifier. Comprehensive evaluations using metrics like accuracy, root mean square error, and area under the curve further highlight the method’s superior performance. Compared to other techniques, this approach demonstrated a notably high recognition rate, advancing the state of the art in image classification. The proposed approach differs from prior works by uniquely combining hand-crafted (SIFT, Haralick) and deep learning (VGG19) features, optimizing feature selection with K-means and LPP, and achieving 99.13% accuracy on complex, noisy datasets, outperforming traditional methods.