HYBRID DOMAIN BASED STEGANOGRAPHY USING BPS, LSB AND IWT

Transcription

1 92 Chapter 6 HYBRID DOMAIN BASED STEGANOGRAPHY USING BPS, LSB AND IWT Table of contents Particulars Page No. 6.1 Introduction HSBLI Embedding Model 6.3 HSBLI Extraction Model. 6.4 Algorithms of HSBLI. 6.5 Performance Analysis Summary

2 Introduction The Steganography is used in covert communication by embedding secret message into cover object. In this chapter, Hybrid domain based Steganography using BPS, LSB and IWT (HSBLI) is proposed. The new concept of grouping pixels of cover image and payload with intensity values less than 128 and more than or equal to 128 are used for LSB and BPS embedding, which results in high values of PSNR and capacity with more security to the payload. The payload is segmented into two equal parts say part 1 and part 2. The pixels of payload (PL) Part I intensity values with less than 128 are considered and the square root is applied to convert 8 bits to 4 bits length. The 4 bits of payload Part I is embedded into cover image pixels with intensity values less than 128 using the LSB replacement method to generate LSB stego object. The cover and PL Part I with intensity values more than or equal to 128 are considered and Bit Plane Slicing (BPS) is employed to convert gray scale into binary images. The 4 - LSB planes of cover image are replaced by 4 - MSB planes of PL Part 1 to generate BPS stego object. The IWT is applied to an intermediate stego object, which is obtained by merging BPS and LSB stego objects. The PL Part 2 is embedded into an LL sub band of intermediate stego object to obtain final stego object. The inverse IWT is used on stego object to generate stego image.

3 HSBLI Embedding model The proposed block diagram of HSBLI embedding of payload is as shown in Figure 6.1. The payload is more secure in the proposed model since (i) The BPS embedding technique is used for cover and payload image pixel intensity values more than or equal to 128. (ii) The LSB embedding technique is used with square root on payload intensity values less than 128 and cover image intensity values less than 128. (iii) The IWT is applied on both BPS stego object and LSB stego object. (iv)the certain portion of spatial domain PL is embedded into the coefficient of transform domain stego object Payload (PL) The gray scale image of suitable sizes with different formats considered to test the performance of an algorithm Cover Image (CI) The cover image can either be color or grayscale image, provided color image is converted into a grayscale image, before the secret information is hidden. The cover images of any size and formats are considered Segmentation The payload is equally segmented into two parts say Payload (PL) part1 and Payload (PL) part 2.

4 95 Payload Image PL Part-2 PL Part-1 Cover Image Pixel 128 Pixel < 128 Pixel < 128 Pixel 128 SQRT (Pixel) LSB Embed BPS Embed BPS Stego Object LSB Stego Object Image Merge Intermediate Stego Object Integer Wavelet Transform LL Band IWT Embed Final Stego Object Inverse IWT Stego Image Figure 6.1 Proposed HSBLI embedding Model.

5 BPS Embedding The gray scale image is represented by pixels with intensity values vary between zero and 255 with each pixel represented by eight bits. In Bit Plane Slicing (BPS), the eight bits of a pixel are split into its constituent binary planes varying from MSB to LSB. For example, the gray scale matrix of size 4*4 is converted into eight matrices of single bit planes using BPS is shown in Figure 6.2. The intensity values of payload part 1 and the cover image are observed and if the pixel intensity values are more than or equal 128, then BPS embedding is used. The intensity values more than or equal 128 of both PL part 1 and cover image are converted into eight bit planes using BPS. The 4 - LSB bit planes of cover image are replaced with 4 - MSB bit planes of PL part 1 to generate BPS stego object. The eight bit planes of the BPS stego object are merged to obtain single BPS stego object with 8 bits per pixels (a) (b) (c) (d) (e) (f) (g) (h) (i) Figure 6.2 (a). Original matrix, (b)-(i) Constituent bit planes of (a)

6 97 Bit-Plane 0 (a) Original image (b) Bit plane 0 (MSB) (c) Bit plane 1 (d) Bit plane 2 (e) Bit plane 3 (f) Bit plane 4 (g) Bit plane 5 (h) Bit plane 6 (i) Bit plane 7 (LSB) Figure 6.3 BPS of an image The BPS is applied on image Pepeer.jpeg to derive corresponding binarized eight bit planes as shown in Figure 6.3. The MSB bit plane image has significant information of original image compare to other 7 bit plane images. The LSB bit plane images has no information/insignificant information of the original image.

7 LSB Embedding The intensity values of PL part 1 and cover image are observed and if intensity values are less than 128, then consider the pixels for LSB embedding. The square root is applied on each pixel value of PL part 1 to compress eight bits into four bits length. The 4 LSB s of cover image pixels are replaced by 4 - bits of PL part 1 to generate LSB stego object Merge BPS Stego object and LSB Stego object are merged by adding to generate an intermediate stego object IWT Embedding The two dimensional integer wavelet transform is apply on intermediate stego object to obtain four subbands such as LL, LH, HL, and HH. The LL band is the low frequency coefficients and three other bands are the high frequency coefficients. The PL part 2 is embedded into the LL band coefficients using four bit LSB replacement method. The four MSB s of PL part 2 are embedded into four LSB bits of LL band coefficient of intermediate stego object to generate the final stego object. The inverse IWT is used on final stego object to derive stego image. 6.3 HSBLI Extraction Model The embedded secret information is retrieved from the stego image at the destination by applying reverse process of embedding. The extraction of payload at the destination is as shown in Figure 6.4. The

8 99 IWT is used on stego image to obtain four sub bands say LL, LH, HL and HH. The four LSB coefficient of LL band has the MSB bits of PL part 2, hence extract to get PL part 2. The inverse IWT is applied on MSB coefficients of the LL band along with all three sub bands LH, HL and HH to generate an intermediate stego image. The intensity values of intermediate stego image are observed, if values are less than 128 then extract four LSB bits of each pixel and convert into decimal value. Each decimal value is squared to obtain payload part 1. If pixel values are more than or equal to 128, then apply BPS to generate eight bit planes. The four LSB bit planes are considered to obtain MSB part of PL part1. The matrix of part of PL part 1 of LSB method is merged with the matrix of part of PL part 1. The LSB and BPS extract of PL part 1 are combined to derive final PL part 1. The final PL part 1 image is merged with PL part 2 images to extract the original payload. 6.4 Algorithms of HSBLI Problem definition: The secret information is embedded into cover object using BPS, LSB and IWT to generate stego image for secure communication The objectives are to (i) increase capacity (ii) increase PSNR HSBLI Embedding Algorithm The secret information is embedded into the cover medium using spatial and transform domain methods are given in Table 6.1. The

9 100 proposed steganography algorithm is more secured as three different techniques are used to embed payload along with some part of the payload and cover image are compressed. Stego Image Integer Wavelet Transform LL LH PL Part-2 is extracted using LSB Technique HL HH Inverse IWT Intermediate Stego Image If Pixel < 128 No Yes LSB to extracts PL Part-1 Matrix BPS to extract PL Part-1 matrix PL Part-1 Merge Extracted Payload Image Figure 6.4 Extraction model of HSBLI

10 101 Table 6.1: Proposed HSBLI Embedding Algorithm Input: Cover image and Payload Output: Stego image. 1. Consider cover image and payload. 2. Decompose payload into PL part 1 and PL part Check intensity values of PL part 1 and cover image. 4. If intensity values are less than 128 then apply a square root on PL part 1 to compress the pixel length from eight to four bits. 5. The 4 bits of CI are replaced by 4 bits of PL part 1 to generate LSB stego object. 6. The BPS embedding is used on pixels having intensity values greater than or equal to 128 of PL part 1 and cover image. 7. The 4 - LSB BPS planes of cover image are replaced by 4 - MSB BPS planes of PL part 1 to generate BPS stego object. 8. The intermediate stego object is obtained by merging BPS and LSB stego objects. 9. The IWT is applied on intermediate stego object and consider LL band. 10. The MSB bits of PL part 2 are embedded into four LSB s of LL band coefficient to generate the final stego object. 11. Apply inverse IWT to generate a stego image HSBLI Extracting Algorithm The secret information is extracted from the stego image by adopting reverse process of embedding is given in Table 6.2.

11 102 Table 6.2 Proposed HSBLI Extraction Algorithm Input : Stego image Output : Extracted payload image 1. Read the stego image. 2. The IWT is applied to generate four sub bands and extract four LSB from each coefficients of LL band to obtain MSB bits of PL part The inverse IWT is applied on MSB coefficients of the LL band along with all three sub bands LH, HL and HH to generate an intermediate stego image. 4. If intensity values of the intermediate stego image are less than 128 then extract four LSB s from each pixel and converted to decimal square and square it to decompress to generate pixel length to eight bits of PL part 1 using LSB. 5. If pixel values are more than or equal to 128 then applying BPS to generate eight bit planes 6. The four LSB bit planes of BPS are considered to obtain MSB bit planes of BPS PL part Both LSB and BPS matrices are merged to derive final PL part 1 image. 8. The final PL part 1 image is merged with PL part 2 images to extract the original payload.

12 Performance Analysis The cover images such as peppers, house, old image and goldhill shown in Figure 6.5 and payload images of different sizes and formats are used to test the proposed algorithm. The PL Lena image is embedded into the CI Barbara using proposed algorithm to generate stego image Barbara is shown in Figure 6.6. It is observed that perceptibility of stego image and cover image are same. The quality of stego image is same as the cover image by appearance and also statistical characteristic of both stego and cover image is almost same, since PSNR value is more than 40 db. (a) Peppers128.jpeg (b) House.jpeg (c) Old Image.jpeg (d) Goldhill1.jpeg Figure 6.5 Cover images

13 104 (a) CI Barbara (b) Payload Lena (c) SI Barbara Figure 6.6 CI, PL and SI images The values of PSNR between CI & SI and between PL & EPL for different payload sizes with constant cover image size are tabulated in Table 6.3. The payload image Lena. Jpeg is embedded into cover image Barbara.bmp (512 * 512) to obtain stego image Barbara.bmp. The PSNR of the CI and SI values are decreases as capacity increases where as the values of PSNR between PL and EPL are almost constant with respect to the capacity. The variation of PSNR between CI and SI with capacity is plotted in the Figure 6.7. The PSNR values decrease as capacity increases. The variation of PSNR of CI Barbara.bmp (512*512) and SI and the PSNR between original PL with different image formats having size of 64*64 with EPL are given in Table 6.4. It is observed that the payload formats do not affect the quality of stego image i.e., PSNR between CI and SI having value around 43 db and the PSNR between original PL and EPL having a value of around 31 db are almost constant for different payload image formats. The PSNR value between CI & SI and between PL & EPL for different cover image formats with constant capacity are given in Table 6.5. It is

14 PSNR of CI and SI (db) PSNR of CI and SI 105 seen that different cover image formats have least effect on the quality of stego image as PSNR values are almost constant. The PSNR value between CI & SI and PL & EPL for different cover images are almost constant are tabulated in Table 6.6. Table 6.3: PSNR Variations for different payload sizes with cover Payload Image size[ lena.jpeg ] image Barbara.BMP (512 * 512) PSNR of CI and SI (db) Capacity (bpp) PSNR of PL and EPL(dB) 64 * * * * * Capacity Capacity (bpp) Figure 6.7 PSNR variations with Capacity

15 106 Table 6.4: PSNR Variations for different PL formats with CI Payload Image formats Lena(64 * 64) Barbara.BMP (512 * 512) PSNR of CI and SI(dB) PSNR of PI and EPL (db) PNG JPEG BMP TIFF GIF Table 6.5 Variation of PSNR values for different CI formats with PL Cover Image formats image Lena.jpeg (64 * 64) PSNR of CI and SI(dB) Capacity (bpp) PSNR of PL and EPL (db) BMP TIFF GIF PNG JPEG Cover image (512 * 512) Table 6.6 PSNR values for different cover images Payload (64 * 64) PSNR of CI and SI (db) PSNR of PL and EPL (db) Goldhill1.jpeg Lena.png House.jpeg Lena.png Old Image.jpeg Lena.png Peppers128.jpeg Lena.png Barbara.bmp Lena.png

16 107 Table 6.7 Comparison of PSNR for proposed algorithm with existing algorithms Methods Techniques PSNR (db) El Sayed M El Alfy et al., [86] Elham Ghasemi et al., [87] Yedla Dinesh and Addanki Purna Ramesh [88] PVD scheme IWT and Genetic algorithm DWT and BPCS Proposed HSBLI method BPS, LSB and IWT The comparison of proposed Adaptive Steganography based on Spatial and Transform domain Techniques with existing techniques such as El.Sayed M El. Alfy et al., [86], Elham Ghasemi et al., [87], and Yedla Dinesh and Addanki Purna Ramesh [88] given in Table 6.7. It is observed that PSNR values are better in the proposed algorithm as compared to existing algorithms due to eight bits of payload are compressed using square root to convert into four bits to embed into the cover image instead of eight bits. The security to the secret information in the propose algorithm is better since (i) (ii) Spatial and Transform domain concepts are used The part of the payload is compressed using square root before embedding. 6.6 Summary The steganography is a technique to hide secret information into a cover object for secure communication. In this chapter HSBLI algorithm is discussed. The payload is decomposed into two equal parts say Part 1 and Part 2. The PL Part 1 pixels with intensity values less than 128 are square rooted to compress eight bit pixel length to

17 108 four bit pixel length. The four bits of PL Part I are embedded into cover image pixels with intensity values less than 128 using an LSB technique to generate LSB stego object. The intensity values more than or equal to 128 of cover image and PL Part I are converted into bit planes. The LSB planes of cover image are replaced by MSB planes of PL Part 1 to generate BPS stego object. The intermediate stego image is obtained by merging LSB and BPS stego object and IWT is applied. The PL Part 2 is embedded into LL subband of intermediate stego object to generate the final stego object. It is observed that the values PSNR in the proposed algorithm are better as compared to existing algorithms.