Dynamic Increased Capacity Approach Steganography in Spatial Domain

Steganography is paradigm that conceals the secret data into some digital medium in such a way so its existence cannot be noticed by human perception [1-9], see the study [10] for a recent review. With the rapid advancement and use of internet as communication channel, steganography has become an important and valuable tool for secure communication between sender and recipient without the snooping of third person. At the beginning the most known processes of steganography, named as substitution and matching processes, were used for secret communication and many other applications like watermarking, document tracking, document authentication etc. These conventional methods are static approaches having fixed capacity in term of embedding rate. The embedding rate was one bit per pixel for the grayscale images and three bits per pixel for the color images [1-3]. Maheswari and Hemanth [7] utilized Fresnelet transform for frequency domain QR code-based image steganography. Tang et al. [8] proposed a reversible and adaptive image steganographic algorithm based on interpolation. The researches [11-20] present the methods in which least significant bit works and in many methods experiments are performed on different levels of least significant bit position. Luo et al. [21] presents the modified form of LSB substitution by using a secret key. Fridrich [22] and Mielikainen [23] improves the performance of the LSB matching by decreasing the change per pixel. There have been various techniques proposed for steganography. We mention some relevant research techniques here and refer to the study [10] for a comprehensive review of other techniques used in steganography. Subhedar and Mankar [24] utilized wavelet transform and QR factorization from matrix analysis to utilize the redundancy inherent in digital images. Chen et al. [25] utilized principle of the error correction coding channel for devising a data hiding system based on spatial domain least significant bit (LSB) steganography [26], and a capacity analysis model for a MIMO-OFDM coding channel. Yeh et al. [27] considered wavelet bit-plane for steganography for compressed images. Chandramouli and Memon [28] used Shearlet transform and bidiagonal singular value decomposition (SVD) for transparent image watermarking.

1.png

Figure 1. Bit plane view of the gray scale Lena image. As we move from left to right, we move from least to most significant bit plane representation

2.png

Figure 2. Effect of inconsistency/disturbance in bit plane when static steganography is applied. From left to right: Clear least bit image, and after 50% and 100% static embedding

There are certain problems with the static steganographic approaches. Static Approaches typically create specific patterns in statistical features of the images that makes the syego-system easily breakable. Static methodologies create disturbance in the correlation of the bit planes of the images, creates static pattern in the histogram of the image e.g. pair of values (POV) [29-31] in case of image substitution steganography. Figure 1 shows the bit plane view of the gray scale Lena image. All these bit planes have correlation with each other. As we move from left to right, we are actually moving from least to most significant bit plane. Although any change in least bit planes does not affect the overall visual appearance of the image, but bit plane slice view will clearly show the effect of disturbance in the corresponding bit plane. Figure 2 shows the effect of disturbance/inconsistency in the correlation of the image bit plane when we apply static steganography methods. Left image shows the clear least bit plane of the image. Central image shows the effect of static embedding methodology after 50% embedding rate and right image shows after 100% embedding rate. Static approaches have fixed capacity in terms of embedding rate. This is a problem because embedding capacity for static methodology depends on the size of the image.

For a strong and unbreakable steganography systems, the methodology of the embedding phase should be design in such a way so that the statistical features of the image should not create repeated patterns. In this paper, a new dynamic approach for the grayscale and color images has been presented. The novel method has increased the embedding rate up to 4 bits per pixel for the grayscale images and up to 12 bits per pixel for the color images. The major challenges to design a new approach includes,

• dynamic embedding methodology,
• increased embedding rate,
• and achieve above said challenges with least changes in image quality and security analysis measures.

The testing criteria or tools used for the evaluation of the new proposed approach will be based on Image Quality Measures, Security Analysis, Histogram Analysis and Bit Plane Analysis [32-40].

Rest of the paper is arranged as follows. Section 2 presents the proposed dynamic increase capacity-based method. Section 3 provides detailed experimental results and discussion. In Section 4 we conclude the paper.

In this section, we investigate the performance of the proposed algorithm by analyzing some benchmark images. These benchmark images include five grayscale and five color images. We analyze all experiments that are carried out on a database of ten images with the size of 512 x 512. As shown in Figure 4, we number these images from one to ten (according to the order from the left-to-right and top-to-bottom).

4.png

Figure 4. Set of test images – we refer to image 1 to image 10, from left-to-right and top-to-bottom. Note that we chose 5 gray scale and 5 color images representing various image contents, flat regions, texture, and strong edges

Table 2. MSE values of the ten test steganography images after different embedding rates

We obtain the steganography images by embedding different embedding rate with the proposed approach. The quality of steganography images is evaluated by using frequently used full-reference image quality assessment methods, such as mean square error (MSE) [6] value and peak signal-to-noise ratio (PSNR) value [6]. Objective performances for ten - five grayscale and five color (RGB - red, green, blue - channels), images are given in Table 2 and Table 3. Less MSE and higher PSNR (dB) values are indicative of better performance. We chose these ten test images that contain various objects and texture characteristics to show the advantages of the proposed dynamic increased capacity approach. We obtained similar results on other standard test images from the image processing literature. Obtained values in Table 2 and Table 3 show that the proposed technique produces very good performance for both the grayscale and color images. The statistical changes are least so that the differences between the original cover and steganography images are undetectable. This assertion is supported by lower MSE values and higher PSNR (dB) values for various test images. We also note that even in the high content texture images such as the Baboon (images 2, 8) we obtained lower MSE values – 0.1067, 0.0969 respectively for gray and color images – at the higher embedding 133712 bits.

Table 3. PSNR (dB) values of the ten test steganography images after different embedding rates

3.1 Bit plane analysis

Bit plane analysis checks the visual inconsistency due to any change in the image pixel value. For the most popular methods including LSB substitution [1], LSB matching [2, 3], GLM method [5] etc. the visual inconsistency is clearly visible in least bit planes and for the robust increased capacity image steganography scheme, least two-bit planes show clear change. The projected scheme of steganography has the advantage over these methods that the bit planes does not show the visual inconsistency in the bit planes.

5a.png

5b.png

Figure 5. Bit plane analysis for the 5 gray scale test images with various embedding bits – 33472, 66832, 100296, 133712 with our dynamic capacity approach steganography. We show for each test image, bit plane images after these four embedding bits. We see that the final bit plane image does not show any visual inconsistencies

6a.png

6b.png

6c.png

6d.png

6e.png

Figure 6. Bit plane analysis for the 5 color test images with various embedding bits – 33472, 66832, 100296, 133712 with our dynamic capacity approach steganography. We show for each test image, bit plane images after these four embedding bits in each color channels

This is since, in the proposed increased capacity approach, the message bits are embedded dynamically. Figure 5 shows the bit planes of images after embedding different bits in the test images from Figure 4. For color images three-bit planes, for red channel, blue channel and green channels, are shown in Figure 6. As can be seen, the proposed approach obtains no visual inconsistencies indicated other models (compare Figure 2) that are manifestation of static steganography methods.

Also, the final bit plane image in both gray scale and color images show that the image features are kept intact, and no artifacts are observed in the resultant images.

3.2 Histogram analysis

Histogram of an image shows the graphical distribution of the picture element value. In image steganography, histogram analysis is used as a steganography analysis tool for various steganography schemes. For LSB substitution [1], LSB matching [2, 3], GLM method [5] etc., histogram creates the specific pattern. For the projected scheme, as the embedding is in dynamic style, so histogram cannot be used as the steganography analysis tool. Histograms of the images after embedding different bits using test images of Figure 1 are shown in Figure 7. For the color images, three channels-based histograms are shown separately in Figure 8. We notice that the histograms do not show any sequenced peaks or discernible patterns that are associated with other related models.

7.png

Figure 7. Histogram analysis for the 5 gray scale test images with various embedding bits – 33472, 66832, 100296, 133712 with our dynamic capacity approach steganography. We show for each test image, bit plane images after these four embedding bits

8a.png

8b.png

Figure 8. Histogram analysis for the 5 gray scale test images with various embedding bits – 33472, 66832, 100296, 133712 with our dynamic capacity approach steganography. We show for each test image, bit plane images after these four embedding bits

3.3 Comparative analysis

Comparative analysis of the projected method with the LSB substitution [1], LSB matching [2, 3], increased capacity in spatial domain [4] and GLM [5] methods is summarized in terms of bit plane and histogram analysis as well as static versus dynamic, bits per pixel in Table 4. It is clear from the Table 4 that new proposed dynamic increased capacity approach is best in terms of embedding rate, visual analysis and histogram analysis.

From experimental results it can be seen that the proposed dynamic approach obtained better results in terms of lower MSE and higher PSNR (dB) indicating the resultant images are indistinguishable from original images. Compared to other methods, our proposed approach also utilized 4 bits per pixel, whereas the previous models at the top use 3 in the case of color images. In terms of histogram analysis, we see that our proposed method is devoid of artificial structures and the corresponding histograms show no discernible patterns whereas the previous models can create either sequenced peaks or maxima/minima changes.

Table 4. Comparative analysis of the projected method

Further, the proposed approach is dynamic in contrast to other static methods for steganography. The least bit planes do not show visible changes, and this is advantageous in for example in medical images category wherein drastic pixel changes are undesirable [37]. Finally, the proposed approach is shown to work on color images that are extensions of gray scale models with increased capacity in the spatial domain. Extending this approach to handle multichannel and multimodal imagery data is an interesting framework that requires further consideration.