In this paper, we evaluate the readability of optically written watermarking from an image compressed by JPEG. We previously proposed an optical watermarking technique that can protect the portrait rights of real objects. It produces a watermarking pattern in the illumination light by modulating color differences. The illumination light that contains such watermarking is pro-jected onto an object. An image of the object taken by a camera contains the same watermarking, which can be extracted by image processing. Therefore, this technique can protect the portrait rights of real objects. We conducted simulations of capturing an object image illuminated by watermarked light, compressing it by JPEG, and reading embedded information from the decoded image. The simulation results showed that the accuracy in reading out embedded information decreases when captured images are compressed. However, for medium-level or low compression rates 100% accuracy can be expected by using the error correction technique.
The distribution of digital images is increasing as the Internet becomes broadband and cloud services become popular. This has increased the importance of protecting the copyright of digital images distributed on the Internet because digital images can easily be copied in a way such that they are exactly like the original.
Digital watermarking is an effective way of protecting digital image copyrights from being illegally copied and various digital watermarking techniques for digital images have been developed [
Digital watermarking has also been used in printed images, where digital watermarking is embedded in the digital data before the images are printed [
However, conventional digital watermarking rests on the premise that people who want to protect the copyrights of their digital content, i.e., content creators or content providers, have the original digital data and they can embed watermarking in the original digital data by digital processing.
However, this premise does not always hold. For example, it does not hold in the case that a person took a photo of a painting at a museum with a digital camera. Since recent digital cameras are highly advanced, captured images have very high levels of quality and if the painting is invaluable as a portrait, the captured image of the painting also has high value. Therefore, the portrait rights of such paintings should be protected. However, images captured with digital cameras do not have watermarking in this case because they have been captured by visitors to museums who are not interested in protecting portrait rights.
We previously proposed a technique that could prevent the copyright or portrait rights of images of real objects captured with a camera [
In this paper, we describe the effects of image compression by JPEG on the readability of the embedded information. The motivation of this study was that image data captured with a digital camera are usually compressed and stored in the camera’s memory and that the image compression may cause the embedded watermarking data to change or disappear. We conducted simulations to evaluate the effects of the image compression and in this paper present the obtained results. This template, created in MS Word 2007, provides authors with most of the formatting specifications needed for preparing electronic versions of their papers. All standard paper components have been specified for three reasons: 1) ease of use when formatting individual papers, 2) automatic compliance to electronic requirements that facilitate the concurrent or later production of electronic products, and 3) conformity of style throughout a journal paper. Margins, column widths, line spacing, and type styles are built-in; examples of the type styles are provided throughout this document and are identified in italic type, within parentheses, following the example. Some components, such as multi-leveled equations, graphics, and tables are not prescribed, although the various table text styles are provided. The formatter will need to create these components, incorporating the applicable criteria that follow.
There are various ways of producing optical watermarking;
After setting HCs for all blocks, the frequency domain data for each block is converted into block image data in the space domain by inverse discrete cosine transformation (i-DCT), and all block images are combined into one image. The
Y is a set constant value in the whole image area. This value gives averaged brightness. The Cr component is set to zero in the whole image area. Finally, the YCbCr signal is converted to an RGB signal, input to a space light modulator (SLM), and changed to illumination light that illuminates real objects such as paintings. We were able to use a commercial projector as an SLM for this purpose.
The captured image of the real object illuminated by the illumination light also has a high-frequency pattern in the Cb component image; that is, the captured image also contains watermarking.
The watermarking pattern in the light and in the captured image cannot be seen by the human visual system because it is modulated at the highest frequency and the amplitude of modulation is small.
After the compressed image is decoded it is converted into a YCbCr signal, and then the Cb component is divided into blocks. The pixel data in each Cb component block is then converted into data in the frequency domain by discrete cosine transformation (DCT). Finally, the embedded data are read out by
checking the sign of the frequency component in which binary data are embedded for each block.
We evaluated the effect of image compression on accuracy in reading out embedded watermarking by simulating the procedure of capturing object images, compressing the images, and reading out the embedded information. We used JPEG to compress the images.
The image data, I(x,y), of an object captured with a camera is proportional to the brightness of the object surface. The brightness, B(x,y), is obtained as a product of the reflectance of the object surface, R(x,y), and the luminance of the projected light at the object surface, L(x,y), as
B ( x , y ) = R ( x , y ) { L ( x , y ) + L 0 } (1)
where L0 is bias luminance such as that produced by room light.
As objects we used standard images that had 512 × 512 pixels, as shown in
We first generated the data of the Cb component in the frequency domain for L(x,y) in Equation (1), as shown in
block. The same “1” and “0” numbers were randomly embedded. The magnitude of HC in the original data was changed from one to 10 as an experimental parameter, while Y, Cr, and L0 were set to constant values of 200, 0, and 40. These values were the gray levels of image data whose maximum was 255.
The captured image, I(x,y), is proportional to B(x,y). However, according to sampling theory, in order for I(x,y) to have the highest frequency component that B(x,y) has, I(x,y) needs to have twice as many pixels in the vertical and horizontal directions as B(x,y). Therefore, we produced I(x,y) by doubling the number of B(x,y) pixels vertically and horizontally. Therefore, the image had 1024 × 1024 pixels.
We used JPEG as an image compression algorithm. JPEG consists of a quantizing process and Huffman encoding. Since Huffman encoding is a reversible process, we simulated only the quantization process. In the JPEG quantization process, frequency components in block data in the frequency domain are divided by certain numerical values predetermined for each frequency component. These divisors are obtained by dividing the values in the quantization table by numerical values called scale factors.
A decoded image was divided into 4096 (64 × 64) blocks. For each block, Cb components were converted into data in the frequency domain by the DCT.
Embedded data were read out by checking the sign of the (7,7) frequency component of the Cb for each block. The accuracy with which data was read out was evaluated on the basis of the percentage of data that was correctly read out from 4096 binary data.
It is desirable for the accuracy to reach 100% for HC under 10 because at that HC level watermarking is almost invisible when Cb is modulated [
In cases where an error can be expected to occur when embedded information is read out, we can use the error correction technique. In fact we used this technique in a previous study where we modulated the luminance [
16 | 11 | 10 | 16 | 24 | 40 | 51 | 61 |
---|---|---|---|---|---|---|---|
12 | 12 | 14 | 19 | 26 | 58 | 60 | 55 |
14 | 13 | 16 | 24 | 40 | 57 | 69 | 56, |
14 | 17 | 22 | 29 | 51 | 87 | 80 | 62, |
18 | 22 | 37 | 56 | 68 | 109 | 103 | 77 |
24 | 35 | 55 | 64 | 81 | 104 | 113 | 92 |
49 | 64 | 78 | 87 | 103 | 121 | 120 | 101 |
72 | 92 | 95 | 98 | 112 | 100 | 103 | 99 |
100% accuracy in reading out embedded information even if it had been around 80% when error correction was not used. Therefore, we expected that in this study 100% accuracy could be achieved by using the error correction technique except for cases when the scale factor was one. A scale factor of one results in high compression rate and poor image quality. A scale factor of five results in medium level compression rate and medium level image quality, and a scale factor of 10 results in low compression rate and high image quality. Therefore, this technique can be used for medium-level and high quality images.
In this study, we evaluated the accuracy of reading out optically written watermarking from a captured image compressed by JPEG. We conducted simulations of capturing an image of an object illuminated by the watermarked light, compressing it by JPEG, and reading out the embedded information from the decoded image. The simulation results showed that accuracy in reading out embedded information decreases when the captured image is compressed, but that for medium-level or low compression rates 100% accuracy can be expected by using the error correction technique.
This study was supported by the Japan Society for the Promotion of Science (JSPS) under Research Institute Grant No. 16H02820.
Uehira, K. and Unno, H. (2018) Effects of JPEG Compression on Reading Optical Watermarking Embedded by Using Color-Difference Mo- dulation. Journal of Computer and Communications, 6, 56-64. https://doi.org/10.4236/jcc.2018.61006