Development of Efficient Watermarking Techniques for Document Images

Abstract

Watermarking is used in copyright protection and authentication of multimedia data such as text, image, audio , video and document images. Among these media, document images exhibit peculiar characteristics such as binary nature, block/line/word patterning, and clear separation between foreground and background areas. As a result, watermarking techniques are required to address these characteristics of document images. In this thesis, four watermarking techniques for document images are proposed. The first technique is based on zero watermarking approach. In this technique, Lifting Wavelet Transform is applied on the cover document image to generate its subbands. These subbands are further divided into non-overlapping blocks of same size. A zero watermark is generated by using the features of these blocks. This watermark is combined with the key watermark to create a meaningful watermark. This meaningful watermark is stored with a trusted authority. On the receiver side, a watermark is generated using the received image and compared with the meaningful watermark to check the authenticity of the image. The second watermarking technique is based on hashing and Zero Width Characters. Hash value of the input image is generated by applying Message Digest 5 on it. A lookup table is used to translate this hash value into Zero Width Characters. Watermark is also translated into Zero Width Characters before embedding in the cover image. The Zero Width Characters are embedded at the end of every sentence. At the receiver side, watermark and hash values are generated from Zero Width Characters extracted from the watermarked image. Watermark value is used for the protection purpose while hash value is used to authenticate the watermarked image. Third technique is developed on the basis of Markov Matrix and Entropy to detect tamper in the text images. Entropy of every sentence of document image is calculated and Markov Matrix is generated by using the occurrences of characters of the cover image. Entropy and character patterns values are converted into Zero Width Characters by using a lookup table. Zero Width Character of entropy of each sentence is embedded at the end of every sentence after terminator. Patterns of the Zero Width Characters are embedded in the end of the text of the cover image. On receiver side, Zero Width Characters are extracted from the watermarked image and converted to numerical form using the same lookup table. Entropy of every sentence and character patterns are recalculated and compared with extracted values for tamper detection. The fourth technique is based on Discrete Curvelet Transform and Discrete Cosine Transform to provide the protection of document images. Discrete Curvelet Transform generates the six layers of the document image. Sixth layer is selected for the embedding process. Discrete Cosine Transform is applied on the 8 × 8 blocks of the sixth layer and the required number of discrete cosine coefficients are selected in zig zag pattern to embed the watermark bits. Watermark bits are extracted at the receiver side by using the steps used in embedding in the reverse order. In extraction process, watermarked image as well as cover image are required. Imperceptibility and robustness parameters are evaluated to measure resistivity of the watermarked image against image processing attacks. Experimental results demonstrate the efficiency of the proposed techniques. Comparisons with existing document image watermarking techniques show the effectiveness of the proposed techniques.