Research Of Feature Extraction On Digital Watermarking Against Geometric Transforms

With the continuous development of image processing technology, the characteristics ofanti-geometric attacks plays an increasingly important role in the field of image processing.Itis widely used in pattern recognition,3D modeling,image registration and other fields.Image features generally refers to those prominent elements compared to its surroundingneighborhood. Image features are generally extracted from grayscale, color, texture. Thecurrent feature extraction are resistant to operation such as filtering, compression, but to thosegeometric transformation like rotation, scaling often extract completely differentcharacteristics compare to the characteristics extracted from the source image. In fact, scaling,rotating does not change image content. This paper seeks a feature extraction algorithm whichtolerate the geometric transformation. The algorithm try to extract the characteristics of theimage content through extracting feature points, and provide a reference point for the localwatermark embedding, and ultimately constructed the watermark embedding area.At present, the image feature extraction has the following ways: points extraction,region extraction; the point of extraction has been the mainstream because of the largenumber of features and representative of the image content. Algorithms mentioned in thispaper are based on the point of extraction, the following is the work of the paper:(1) Introduced the representative feature point extraction algorithm such as SIFT, Harris,Harris-Laplace, DRSIF.(2) Proposed an extraction algorithm based on the Harris.(3) Base on the watermark area construction algorithm based on SIFT, proposed a algorithmbase on Harris-Laplace for constructing watermark embedding area and the effect is betterthan the former.(4) According to the shortcomings of the DRSIF feature extraction algorithm, a new algorithmbased on DRSIF and salient region selection are proposed.(5) Compared the proposed algorithms with the algorithm based on SIFT from the accuracyand algorithm execution time, and made a summary. It can be seen from the experimental two watermark region construction algorithmproposed by this paper obtain a better accuracy for geometric attacks.