This paper presents a feature-based robust watermarking scheme. The proposed method is to discover the feature from original image against SIFT technique and then proceed with image normalization based on local feature region (LFR) to create one normalized image with fixative size and direction. Afterwards, proceed with the embedding and detection process toward the normalized image. In this proposed method, it can counteract not only the global transformation but local transformation. According to the experimental data, the proposed method is better than Lu’s method; furthermore, the existence of watermark can also be verified by only using the statistic method without any requirement of original image. Therefore, our proposed method is an efficient and effective watermarking method.