关键词: 尺度不变特征变换, 局部特征区域, 嵌入强度, Watson人类视觉模型

Abstract: To solve the problems that the digital watermark information are vulnerable to geometric attacks and the balance between invisibility and robustness of watermarking algorithm, a digital watermarking algorithm was proposed based on Scale Invariant Feature Transform (SIFT) for image local feature points. Based on SIFT algorithm local feature points, circular Local Feature Area (LFA) that met certain conditions were found. After the LFA regularization, the watermarking was embedded in intermediate frequency coefficient of LFA Discrete Cosine Transform (DCT) domain. Among them, considering the influence of the watermark information on the image quality, the embedding strength was dynamically adjusted according to the Watson human visual model. The experimental results show that the Peak Signal-to-Noise Ratio (PSNR) and the similarity of watermark get higher. The algorithm guarantees better watermark invisibility and robustness performance under certain geometric attacks.

Key words: Scale Invariant Feature Transform (SIFT), Local Feature Area (LFA), embedding strength, Watson human visual model