Effective Digital Image Watermark Detection Technique based on Tamura Texture Features

One technique for limiting unauthorized changes to digital images is image watermarking. However, there are a variety of challenges with these techniques, including a poor detection accuracy rate, high processing costs, and resource usage. Due to these challenges, finding watermark in the images is important issue. Therefore, we proposed an effective watermark detection technique by combining the best texture features (Tamura) and classifier (Canonical Correlation Analysis (CCA)). The proposed watermark detection method consists of three steps: image conversion, feature extraction, and classification. The RGB images are first transformed into grayscale versions. In the feature extraction step, the Tamura is extracted from gray-scale images. Finally, Canonical Correlation Analysis (CCA), Random Forest (RF), and Adaboost perform the classification. We use the publicly available benchmark dataset VOC2012 to evaluate the performance of the proposed technique. The results of the experiment show that the Tamura texture feature is the optimal texture feature to utilize in detecting watermarks because it had the best accuracy across all classifiers. It also demonstrates that the CCA classifier fared better in terms of accuracy than the other two. The second-best classifier belongs to Adaboost. However, among the classifiers, the RF classifier exhibited the lowest accuracy. The proposed Tamura based CCA method (Tamura-CCA) is compared to other state-of-the-art techniques in terms of feature dimension and detection accuracy. The proposed method performs better than other existing techniques. The Tamura-CCA has an accuracy rate of 98%. The proposed method has the benefits of the ability to locate watermarked images in every database and not needing a particular type or procedure for the watermark embedding.