Classification of streaming platform images using local binary patterns and machine learning algorithms

This research has developed an effective multi-class classification model for images from streaming platforms. Texture features were extracted from the images and used with machine learning algorithms. Two datasets were employed: ’mosaics’ comprising 153,488 images across 14 classes and ’descriptors’ containing 33,471 images across 11 classes. All images had a resolution of 1280 × 720 pixels. The local binary pattern (LBP) technique encoded the local texture structure into 59-element feature vectors for each image. Ten algorithms were trained and evaluated on these vectors for each dataset, including support vector machines (SVM) with linear, polynomial, and Gaussian kernels at various scales, as well as ensemble methods like boosted trees, bagging, discriminant analysis, and k-nearest neighbors in subspaces. Training and validation were done via 30 random splits to mitigate bias. Performance metrics like accuracy, sensitivity, specificity, precision, and F1-score were computed per class. The SVM classifier achieved top mean performance: 0.998952 accuracy, 0.992528 sensitivity, 0.999438 specificity, 0.988132 precision, and 0.990280 F1-score. The results validate the proposed LBP feature extraction and machine learning methodology for effectively classifying images across streaming platforms.