Adaptive data hiding based on VQ compressed image

  • Slides: 16
Download presentation
Adaptive data hiding based on VQ compressed image Authors: Wei-Chang Du and Wei-Jun Hsu

Adaptive data hiding based on VQ compressed image Authors: Wei-Chang Du and Wei-Jun Hsu Source: IEE Proceedings-Vision, Image and Signal Processing, Vol. 150, No. 4, August 2003, pp. 233 -238 Speaker: Chi-Nan Lin Date: 2004/04/21 1

Outline n Introduction n Image coding using VQ n Data hiding on VQ n

Outline n Introduction n Image coding using VQ n Data hiding on VQ n Experimental results n Conclusions 2

Introduction (1) n Two fundamental requirements on data hiding in an image: n n

Introduction (1) n Two fundamental requirements on data hiding in an image: n n The distortion of cover image should be minimized. The hiding capacity should be large enough. Usually the above two requirements are a trade off of each other. Earlier study: n On spatial domain. n ex. to hide data in the LSB bit. 3

Introduction (2) n Recent study: n To hide data in compressed images. n n

Introduction (2) n Recent study: n To hide data in compressed images. n n n -->in palette-based images (ex. , GIF image). -->in frequency-domain images (ex. , JPEG image) -->in VQ compressed images. 4

Image coding using VQ(1) VQ Encoder 5

Image coding using VQ(1) VQ Encoder 5

Image coding using VQ(2) VQ Decoder 6

Image coding using VQ(2) VQ Decoder 6

Data hiding on VQ (1) n Pre-processing: n n n Secret data should be

Data hiding on VQ (1) n Pre-processing: n n n Secret data should be compressed to save space. Secret data need to be encrypted for security (use DES, RSA etc. ). Fixed embedding method: n n Data to hide (embed) in the LSB of indices on VQ compressed format. Codebook was divided into two subcodebooks. 7

Data hiding on VQ (2) n MGLE (mean grey-level embedding) or PNNE (pairwise nearest-neighbour

Data hiding on VQ (2) n MGLE (mean grey-level embedding) or PNNE (pairwise nearest-neighbour embedding) the (MGLE) was used to divide (PNNE) codebook (into two sub-codebooks). 8

Data hiding on VQ (3) (the secret data) Øcodeword Yi and Yi+1 are very

Data hiding on VQ (3) (the secret data) Øcodeword Yi and Yi+1 are very similar to each other. Øsecret bit stream help decide which sub-codebook to use. embedding method: n Fixed 9

Data hiding on VQ (4) n Fixed embedding method: n Two main drawbacks: n

Data hiding on VQ (4) n Fixed embedding method: n Two main drawbacks: n n Fixed capacity --> each image block can hide only one bit. Large distortion --> especially when MGLE was (PNNE) used. (MGLE) 10

Data hiding on VQ (5) (MGLE) n (PNNE) Adaptive clustering embedding (ACE) Adaptive embedding

Data hiding on VQ (5) (MGLE) n (PNNE) Adaptive clustering embedding (ACE) Adaptive embedding method: 11

Data hiding on VQ (6) Y 1 Y 2 Y 3 Y 4 Grouping

Data hiding on VQ (6) Y 1 Y 2 Y 3 Y 4 Grouping Y 5 G 0 Y 1 Y 2 G 1 G 2 Y 3 G 3 Y 7 Y 8 Y 6 (ACE) n Y 7 Y 8 Y 4 Y 5 Y 6 unsigned integer value of bit stream, stop grouping Adaptive embedding method: 12

Data hiding on VQ (7) G 0 Y 1 Y 2 G 1 G

Data hiding on VQ (7) G 0 Y 1 Y 2 G 1 G 2 Y 3 G 3 Y 4 Y 5 Y 6 Y 7 Y 8 Codebook n Image encoded by VQ Y 7 Y 5 Y 2 G 3 : ︷ Y 7 Y 8 ︸ G 2 : ︷ Y 4 Y 5 Y 6 ︸ G 0 : ︷ Y 1 Y 2 ︸ Bit stream: (1101)2=13 Y 7(0) Y 8(1) Y 4(00) Y 5(01) Y 6(10) Y 1(0) Y 2(1) After embedding Y 8 Y 6 Y 2 Adaptive embedding method: 13

Experimental Results (1) Original image VQ PSNR: 31. 24 MGLE PNNE ACE Results for

Experimental Results (1) Original image VQ PSNR: 31. 24 MGLE PNNE ACE Results for various embedding methods (capacity = 16 Kbit) 14

Experimental Results (2) 15

Experimental Results (2) 15

Conclusions Based on VQ compressed images. n ACE (Adaptive Clustering Embedding) has higher hiding

Conclusions Based on VQ compressed images. n ACE (Adaptive Clustering Embedding) has higher hiding capacity and better image quality than traditional methods. n It is possible for ACE to be applied to other media in which VQ is applicable. n 16