InceptionV3-Driven Multiclassifier Voting System for Watermark Classification

Watermark text detection and classification are fundamental tasks in computer vision. This study addresses the challenge of classifying watermarked versus non-watermarked text in natural scene images by utilizing the InceptionV3 model for feature extraction. We evaluate the performance of three clas-sifiers— k-Nearest Neighbors (k-NN), Support Vector Machine (SVM), and Random Forest (RF)—in distinguishing between these text types across two distinct datasets. By applying our method to these datasets, we ensure a comprehensive evaluation and assess the robustness of our approach. The results underscore the effectiveness of the InceptionV3 model in capturing relevant features, followed by a comparative analysis of the classifiers based on accuracy, precision, and recall. To further enhance classification performance, a weighted voting decision mechanism is employed, leveraging the strengths of each classifier. The weighted voting accuracy for Dataset1 and Dataset2 are 95.02% and 92.72%, respectively. The weighted voting approach com-bines individual classifier predictions based on their performance metrics, resulting in improved accuracy and more balanced classification results.