160 likes | 174 Views
A Digital Image Watermarking Method Based on Labeled Bisecting Clustering Algorithm. Source : IEICE Transactions on Fundamentals, VOL.E87-A,NO.1 JANUARY 2004 pp.282-285 Authors : Shu-Chuan CHU, John F. RODDICK, Zhe-Ming LU and Jeng-Shyang PAN
E N D
A Digital Image Watermarking Method Based on Labeled Bisecting Clustering Algorithm Source: IEICE Transactions on Fundamentals, VOL.E87-A,NO.1 JANUARY 2004 pp.282-285 Authors:Shu-Chuan CHU, John F. RODDICK, Zhe-Ming LU and Jeng-Shyang PAN Speaker:Chuen-Ko Tsai Date:2004/05/12
Outline • Introduction • The Watermarking Algorithm • Experimental Results
Introduction Privacy key Original image Embedded image Embedding algorithm Watermark Watermark Extracting algorithm Privacy key
A labeled bisecting clustering algorithm • Step 1:The whole training set is viewed a single cluster. Split this cluster into two sub-clusters. One is labeled ‘0’, the other is labeled ‘1’. • Step 2:Pick the cluster Cp that has the largest distortion to split. • Step 3:Find 2 sub-clusters using the basic LGB algorithm (Bitsecting step). • Step 4:Repeat Step 3 Im times and take the split that produces the clustering with the highest overall similarity. Thus, we can obtain two new clusters Ca and Cb.
A labeled bisecting clustering algorithm (cont.) • Step 5:For cluster Ca and Cb, find their neighboring clusters other than each other. If Ca has a nearest neighboring cluster Cc labeled l, and Cb has no neighboring clusters, the Ca is labeled 1-l and Cb is labeled l. Otherwise, if Ca has a nearest neighboring clustering Cc labeled l, and Cb also has a nearest neighboring cluster Cdlabeled m, then Cais labeled 1-l and Cbis labeled 1-m.
A labeled bisecting clustering algorithm (cont.) • Step 6:Repeat steps 2, 3, 4, 5 until the desired number of clusters is reached. • Step 7:Record all cluster labels and centers to form the labeling key Keyl and the final codebook C, respectively.
w h The Original image and watermark The binary watermark image of size 128*128 256-grayscale Lena image of size 512*512
Step 1 Step 2 The whole training set is viewed as a single cluster Cp 0 1 The step for generate the codeword-labeled VQ codebook
Step 3 Ca 0 1 0 1 Cp Cb The step for generate the codeword-labeled VQ codebook (cont.) Step 4 Find 2 sub-clusters using the basic LGB algorithm
Step 5 Cc Ca 0 1 Cb The step for generate the codeword-labeled VQ codebook (cont.) Step 6 Repeat steps 2, 3, 4 and 5 until the desired number of clusters is reached Step 7 Record all cluster labels and centers to form the labeling key Keyl and the final codebook C, respectively 0
A example to describe the embedding process for each input vector