Adaptive data hiding for vector quantization image based
Adaptive data hiding for vector quantization image based on overlapping codeword clustering Author: Yen-Shing Tsai and Piyu Tsia* Source: Information Sciences 181(2011) 3188 -3198 Reporter: Hsin-Tse Lee Date: 2011/12/22
Outline • Vector Quantization(VQ) • Related Work ▪ Chiang and Tsai’s scheme • Proposed Scheme ▪ The proposed overlapping codeword clustering ▪ The embedding procedure ▪ The extraction procedure • Experimental Result • Conclusions 2
Vector Quantization(1/2) Encode: Find Closest codeword w … h … … … Image Codebook Codework i i k … Index table k=wxh 3
Vector Quantization(2/2) Decode: Codebook Look-up table Index i Codework w h i Reconstruction image k Approximate block k=wxh 4
Related Work (1/4) Chiang and Tsai’s scheme: Step 2 Step 3 Pre-procedure: Step 1 : codebook clustering Embedding: Step 2 : codeword mapping and embedding Extracting: Step 3 : according to the index to extract the embedded information Step 1 5
Related Work (2/4) codebook clustering: Sub-clustering rules: 1. Similarity : SED(squared Euclidean distance) 2. Sub-cluster : (1) the order of the overlapping codeword is the same (2) the size of the overlapping sub-codebooks is identical For example: (100, 100) CW 2 (110, 115) CW 3 (95, 120) CW 4 (120, 95) CW 5 (80, 90) CW 6 (130, 85) CW 7 (90, 80) CW 8 (105, 90) Step 1 : Calculate the smallest SED with : Step 2: calculate the summed SED to sub-cluster are cluster into 4 -mem: … CW 1 For 3 -mem: and So cluster overlap in the 4 -mem to 1 -mem 6
Related Work (3/4) Embedding : 5 3 84 78 Secret bits : 11 0 101 index 0 1 0 0 1 1 4 2 2 3 2 5 3 7 4 -mem Case : … 77 78 84 78 87 77 77 … Step 1: find the index of codeword 79 255 Step 2: chose the corresponding codeword Step 3: set to index table Index of codeword =5 4 -mem Embed bits = The replacing index= 7 7 2 84 77 2 -mem Case : 255 Sub-clusters Codebook Similarity to 4 -mem but embed 1 bit Index =3 The replacing index= 2 index table 3 -mem Case : Use two index represent 3 bits by production rules from mapping table Index : (84, 78)=>(84, 77) 7
Related Work (4/4) Extracting: 7 2 84 77 index 0 0 0 1 1 2 2 1 2 3 1. The order of 7 = 4 3 5 7 78 Secret bits : 11 0 101 3. The order of (84, 77) = (1, 2) … 79 84 77 78 87 77 According to the mapping table : … Find the corresponding order and transform to binary 77 2. The order of 2 = 255 Sub-clusters Codebook 8
Propose Scheme (1/4) Mapping table: • Sub-codebook’s size and codeword’s order can determine the secret message. • A codeword can reside in other sub-codebooks at the same time. The propose mapping table: 9
Propose Scheme (2/4) Overlapping codeword clustering : Use the codeword’s similarity and mapping table to partition into any size of sub-codebook. The above codebook clustering example: can be cluster 3 -mem 10
Propose Scheme (3/4) Embedding : index 0 1 2 3 0 0 5 3 84 78 1 1 4 5 7 2 1 4 5 7 79 84 77 78 87 77 78 … 255 5 Secret bits : 110 101 1. The index =5, can embed … 77 4 Set 3 to index table 2 3 2. The index =3, can embed 3 7 84 Set 7 to index table 3. The index =84, can embed index table 255 Sub-clusters Codebook Mapping table: 11
Propose Scheme (4/4) Extracting: 3 7 84 index 0 2 3 0 0 1 1 4 5 7 2 1 4 5 7 79 84 77 78 87 77 … 1 77 index table 78 … 255 4 2 5 3 1. The index =3 of 6 -mem, order =5 Look up mapping table, the embedding secret = 2. The index =7 of 6 -mem, order =3 the corresponding secret = Secret bits: 110 101 10 3. The index =84 of 3 -mem, order =1 the corresponding secret = 255 Sub-clusters Mapping table: Codebook 12
Experimental Result(1/5) Cover images of 512*512 13
Experimental Result(2/5) 9618*(16/6)+528*(11/5)+2268*(8/4)+10*(5/3)+311*(2/2)=31673 14
Experimental Result(3/5) Stego-images 15
Experimental Result(4/5) Relationship between the stego-image quality and hiding capacity 16
Experimental Result(5/5) • In Chiange et al. ’s and Chang and Wu’s schemes, the codewords are clustered into 4 -mem, 3 -mem, 2 -mem and 1 -mem sub-codebook without codeword overlapping. • In Chiang and Tsai’s scheme, the codewords are clustered into 4 -mem, 3 -mem, 2 mem and 1 -mem sub-codebook with codeword overlapping. 17
Conclusions • Large of hiding capacity. 18
Thanks for your listening 19
- Slides: 19