| Multimedia information technology not only has brought us many benefits in our dailylife, it also brought us many security problems. Digital image is easy to be acquire, process,transmit and has big redundancy, so it’s suitable for secretly communication. But thesecharacteristics of digital image may also be used by the terrorists to transmit secret message,and it’s a big threat to our public security and national security.Natural image can reflect the real world. Image which contains different scenes mayhas different characteristics in their texture regions and marginal regions. In recent years,many new steganographic algorithms use many strategies to reduce image quality degradationwhich is caused by steganography and keep some image statistical characteristics unchanged,so they can enhance the security of the algorithm. There are many strategies such as embedsecret message adaptively in image rich textures, or combine coding technology withtraditional data hiding technology. While many new Steganalysis algorithm choose imagepre-processing strategy to suppress the image content, then they extract features from theimage noise residual so the features can be much sensitive to the embedded data.In this paper, we do some research about adaptive steganographic algorithm, filter designand feature selection. Based on research and analysis of traditional steganography algorithms,we propose a pixel block-based adaptive steganographic algorithm with embedding prioritygiven to image texture regions. For steganalysis, we select26kinds of filters to build26algorithms and test their performance. What’s more, we combined natural image spatial andfrequency domain characteristics and proposed three improved algorithms. The contributionsof this thesis can be described as follows:1. We do research and analysis on LSBM\LSBMR\EALSBMR algorithms. Based on therelationship between the least two significant bits of neighboring pixels, we propose a pixelblock-based adaptive steganographic algorithm with embedding priority given to imagetexture regions. In other words, the embedding order is from rich textures to flat regions. Ourcriterion for texture assessment can adaptively allocate image regions for embeddingaccording to the length of the secret message. To deal with possible irregular embeddingblocks, we propose a pixel value modification solution. We have justified this solution mathematically. Experimental results have demonstrated that our algorithm outperforms twoestablished LSB (least significant bit) embedding algorithms and one current edge adaptivesteganographic algorithm in terms of embedding efficiency; meanwhile, our algorithm oftenhas stronger resistance to the two representative steganalysis algorithms.2. We analyzed the algorithm proposed by Shi and select26different filters to improvethe algorithm and test their performance. For the algorithm proposed by Chen which isdesigned for texture images, we proposed three different algorithms to apply it to the naturalimage, the experiment results show that the third one has good performance when detectingLSBM algorithm. |
