Keywords

1 Introduction

Digital watermarking is a method to prevent illegal copying of digital content as it can be copied and edited easily. Digital watermarking can be done in various ways. It can be done in spatial domain using least significant bit (LSB) technique. It can also be done in spectral domain using various transforms such as discrete fourier transform (DFT), discrete cosine transform (DCT), and discrete wavelet transform (DWT). Another method of digital watermarking is based on neural network. Various types of neural network algorithm like back propagation neural network, counter propagation neural network, etc., can be used for it. This method is highly secure because in this method, watermarked image is not sent so it cannot be harmed.

2 Classification of Digital Watermarking Schemes

Various types of watermarking methods are used for the protection of digital data. Some of which are:

2.1 Spatial Domain Watermarking Technique

In spatial domain, watermarking is done in pixel domain. The pixel domain methods have main strengths that they are theoretically simple and have very less computational complexities. Embedding of the watermark into cover image is based on the operations like shifting or replacing of bits. Most commonly used spatial domain watermarking technique is least significant bit technique. In this technique, pixel values of cover image as well as watermark image are converted into binary form. The bits of watermark image replace the least significant bit of cover image and in this way, watermark can be embedded into cover image. Figure 1 shows the framework of the embedding using LSB technique.

Fig. 1
figure 1

Embedding of watermark using LSB technique

Full size image

The extraction is also very simple. Watermark data can be extracted by matching the supposed sample with the received data. At the extractor end, a zero matrix equal to the size of watermark is taken for the purpose of extraction. Each element of zero matrix is converted into binary form as well as watermarked image pixels are also converted into binary form. The least significant bits of watermarked image are replaced by each bit of zero matrix. In this way watermark is retrieved by the extractor. Figure 2 shows the framework of the extraction using LSB technique.

Fig. 2
figure 2

Extraction of watermark using LSB technique

Full size image

In the proposed method, the cover image is of size m x n and the watermark image is of size (m × n)/8. The 8th bit of each pixel of cover image is replaced by each bit of the watermark image. The 8th bit of a binary number has least significance so its effect on the cover image is minimum. In this way watermark is embedded and watermarked image is obtained. The performance will be measured using MSE, peak signal-to-noise ratio (PSNR), and normalized correlation (NC). The process is shown in Fig. 3.

Fig. 3
figure 3

Process of LSB watermarking using 8th bit

Full size image

2.2 Spectral Domain Watermarking Technique

2.2.1 Watermarking Using DCT

The DCT is a very favored transform function used in digital signal processing. DCT can also be applied in pattern recognition, data compression, and image processing.

Figure 4 shows the framework of the embedding using DCT. Digital watermarking can be done by applying DCT on cover image to get transformed coefficients. If cover image coefficient is represented as C a , W i is the corresponding bit of the message data, α denotes watermarking strength, and watermarked coefficient is represented as C aw then coefficients are altered depending upon the stream bits of the message using the equation

Fig. 4
figure 4

Embedding of watermark using DCT

Full size image
$$ C_{aw} = C_{a} (1 + \alpha W_{i} )\, $$
(1)

Figure 5 shows the framework of the extraction using DCT. The extraction can be done in reverse manner. The extracted image can be obtained depending upon the difference between the original DCT coefficients and the watermarked image ones. It can be obtained by the following formula:

Fig. 5
figure 5

Extraction of watermark using DCT

Full size image
$$ W_{i} = \frac{1}{\alpha }(C_{aw} - C_{a} ) $$
(2)

2.2.2 Watermarking Using DWT

Wavelet technique is another significant domain for watermarking. When DWT is applied to an image, it decomposes the image into four significant components which are lower resolution (LL), horizontal (HL), vertical (LH), and diagonal (HH) detail components. Figure 6 shows the framework of the watermark embedding using DWT. Watermarking using DWT can be done by applying DWT on cover image to decompose it into four parts. If cover image coefficient is represented as C a , it is decomposed into four parts, W i is also decomposed into four parts, α represents watermarking strength and watermarked decomposition is represented as C aw then coefficients are altered depending upon the stream bits of the message using equation

Fig. 6
figure 6

Embedding of watermark using DWT

Full size image
$$ C_{aw} = C_{a} (1 + \alpha W_{i} )\, $$
(3)

Figure 7 shows the framework of the extraction using DWT. The extraction can be done in reverse manner. The extraction can be done by subtracting the original DWT coefficients from the watermarked image ones. It can be obtained by the following formula:

Fig. 7
figure 7

Extraction of watermark using DWT

Full size image
$$ W_{i} = \frac{1}{\alpha }(C_{aw} - C_{a} ) $$
(4)

2.2.3 Watermarking Using Back Propagation Neural Network

Digital watermarking can be done using back propagation neural network (BPN). BPN can be used to embed the watermark as well as to extract the watermark.

Embedding of watermark using BPN can be done using following steps:

  • The cover image and watermark image are divided into small fragments of size 2 × 1.

  • A BPN is taken with input layer, one hidden layer, and output layer.

  • The fragments of cover image are supplied as input to the BPN and the network is trained to generate the fragments of watermark image. Weights are adjusted to produce the desired output for the given input.

  • The weights are stored in a file and the cover image with the weight file is sent to the extractor.

The process of watermark embedding is shown in Fig. 8.

Fig. 8
figure 8

Watermark embedding using BPN

Full size image

Extraction of watermark using BPN can be done using following steps:

  • At the extractor end, both files are received (weight file and cover image).

  • The cover image is divided into small fragments of size 2x1.

  • The weights are extracted from the weight file and BPN is reconstructed.

  • With the help of fragments of cover image and trained weights, BPN gives the output same as watermark image.

The process of watermark extraction is shown in Fig. 9.

Fig. 9
figure 9

Watermark extraction using BPN

Full size image

The performance of this technique is also measured for noised image. Various types of noises are used such as Gaussian noise, Poisson noise, salt-and-pepper noise, and speckle noise.

3 Results

Digital watermarking can be done using various techniques. Watermark is embedded in cover image and the embedded image is sent to the receiver. PSNR and NC give the robustness of the technique. The time consumed by different techniques has been also compared in this work. The results obtained are as follows: (Tables 1, 2 and 3) (Figs. 10, 11 and 12).

Table 1 PSNR values for digital watermarking techniques (dB)
Full size table
Table 2 NC values for digital watermarking techniques
Full size table
Table 3 Time consumed in various digital watermarking techniques
Full size table
Fig. 10
figure 10

Graphical representation of PSNR for the proposed technique

Full size image
Fig. 11
figure 11

Graphical representation of NC for the proposed technique

Full size image
Fig. 12
figure 12

Graphical representation of time consumed for the proposed technique

Full size image

4 Conclusion

In this work LSB, DCT, DWT, and BPN are used to embed the watermark with cover image which is being sent to the extractor. The performance has been evaluated using PSNR and NC. On the basis of above results, it is clear that spatial domain is the easiest method but it is less secure. Watermarking using DCT and DWT is more robust. The results of watermarking using BPN are best and it is robust as well as secure technique. But the time consumed in BPN technique is higher than in other techniques.

5 Future Work

This work can be further developed using high security algorithms for embedding and extraction of watermark using full counter propagation neural network, etc.