Abstract：Based on the fundamental principles of feature extraction of PCA and KPCA, an improved method to extract features of non-linear image so as to rebuild image was proposed for detecting embedded watermark image defects. This method decreased greatly the dimension of Core Matrix and kept the information of embedded watermark image effectively. Thus the defects of images could be found out through comparison. The experimental results show that this method enables input data to reduce dimension effectively, shortens computation time and improves detection effect and accuracy. KPCA has a higher performance index and wider range of application than PCA.
2010, Vol. 30 
