Research On SVD In Digital Watermark And Infrared Small Target Preprocessing

Singular value decomposition (SVD) has been widely used in many image processing fields in recent years. Some problems in using SVD in digital watermark and infrared small target preprocessing are researched in this paper.The current state of watermark algorithms using SVD is analyzed. Mistakes are easily found in algorithms which only modify singular values (SVs) to embed watermark. To avoid these mistakes, few algorithms based on modifying singular vector matrix have been proposed. But the robustness of the extracted watermark reduced as these algorithms broke the orthogonality of the singular vector matrix. To solve this problem, and maintain the orthogonality of the singular vector matrix, a new watermark embedding algorithm based on modifying singular vector matrices of host image's sub-blocks using the orthogonal relationship of the trigonometric functions has been proposed in this paper. Experimental results show that the proposed algorithm has higher robustness than traditional ones which modify singular vector matrices of host image's sub-blocks because of the proposed algorithm maintains the orthogonality well. The error ratio of extracted watermark has a two orders of magnitude reduction in salt and pepper noise attack. The same ratio in cropping attack is reduced at least 50%. During more attacks, the proposed algorithm can resist space-domain attacks well, but not well in frequency-domain attacks such as JPEG compression.The reason the proposed algorithm above can not resist frequency-domain attacks is lacking of a frequency-domain embedding algorithm. To solve this problem, a new algorithm which embeds watermark by modifying the low frequency DCT coefficients of sub-blocks has been proposed based on the analyzing of basic principles of JPEG compression. The sub-blocks are chosen by calculating their mean square errors. This algorithm maintains the orthogonality of singular vector matrices of sub-blocks'low frequency DCT coefficients. The robustness against frequency-domain attacks such as JPEG compression improves significantly compared with the proposed algorithm above. For example, in quality 10, the error ratio is about 60% using the above algorithm, but only about 16% using this one in JPEG compression.Base on the advantages of the two proposed algorithms, a combination algorithm has been proposed. Experimental results show that the combination algorithm has equivalent effects against many kinds of attack both in space-domain and frequency-domain compared with MWT&EMD algorithm. Compared with traditional algorithms which modify singular vector matrices of host image's sub-blocks, the robustness in cropping attack and salt and pepper noise attack improves significantly and has equivalent effects against JPEG compression. Singular values represent the energy characteristics of an image, and infrared small target can be considered as energy disturbance point in the image. A new preprocessing algorithm of infrared small target image based on SVD has been proposed. The characteristics of SVD of infrared small target and the relation among the point target's SVs, location, size and intensity have been studied. A differential percentage curve algorithm based on singular value curve prediction and the original singular value curve is proposed to reconstruction infrared small target. Experimental results show that the proposed algorithm can improve image signal noise ratio and suppress background clutter well compared with current algorithms. Compared with the commonly used Butterworth filter, effects are improved in all aspects about twice. Compared with current SVD band-pass filter, effects are improved about at least 50%. This algorithm provides a perfect algorithm which can preprocess image of different contexts under the same framework.