Overview of Proposed Methods

In this study, we try to use information hiding techniques for privacy protection

4

in video surveillance. However, it demands a huge embedding capability to embed the privacy information into the video by traditional information hiding techniques. In this view, we try to use a reversible mapping function that allows mapped values to be controllable in magnitudes. Accordingly, the first method proposed in this study embeds a visible meaningful disguise image selected from the background image into a specific private region in a surveillance video. The second proposed method, which is based on the former one, is designed to protect private motion activities in surveillance videos. This method utilizes image processing techniques to improve the perceptual quality of the mapping result as well. Finally, we modify the first method to propose a third new method for image steganography. Brief descriptions of these methods are given subsequently.

1.3.1 Definitions of terminologies

At first, the definitions of some related terminologies used in this study are defined as follows.

1. Privacy-sensitive image: a privacy-sensitive image is an image which poses danger to privacy probably and needs to be concealed.

2. Background image: a background image is an image portion used to cover part of a privacy-sensitive image in this study.

3. Camouflage image: a camouflage image is an image produced by disguising a privacy-sensitive image to be similar to a background image.

4. Protected image: a protected image is a stego-image produced by embedding some recovery information into a camouflage image.

5. Removed bit: a removed bit is a least significant bit of a prediction-residue value.

5

6. Recovery sequence: a recovery sequence is a sequence which records the location of the privacy-sensitive image and the removed bits.

7. Prediction image: a prediction image is an image produced by estimating a source image.

8. Prediction-residue image: a prediction-residue image is an image obtained by computing the difference of an image and its prediction image.

9. Camouflage process: a camouflage process produces a camouflage image from a prediction-residue image of a privacy-sensitive image and a background image.

10. Embedding process: an embedding process embeds data into an image.

11. Recovered image: a recovered image is an image produced by removing embedded data from a stego-image.

12. Extraction process: an extraction process extracts hidden data from an image.

13. Recovery process: a recovery process recovers the original cover image from a stego-image.

14. Lossless recovery process: a lossless recovery process removes the embedded data from a stego-image and retrieves the original cover image without distortion.

15. Image steganography: image steganography is the technique to imperceptibly embed information into images for particular purposes such as cover communication and authentication.

16. Secret image: a secret image is an important image that should be protected properly and not be revealed to unauthorized people.

17. Target image: a target image is an image which is provided by the user and used to produce a camouflage image.

6

1.3.2 Brief description of proposed method for protection of selected private regions in surveillance videos using a reversible prediction-based mapping

In this study, we design a new method to protect privacy-sensitive images using a prediction-based mapping function proposed in this study, which is formed by a combination of two functions  a modified version of a prediction function coming from the JPEG-LS standard [28] and a reversible one-to-one mapping function proposed by Liu and Tsai [16]. First, we use the first function to generate a prediction-residue image with small pixel values. Then, we map the residue image and a pre-selected background image together into a third image, called a camouflage image, using the second function proposed by Liu and Tsai [16]. The resulting camouflage image is similar to the selected background image and it is hard to tell the difference between them by human eyes. At last, in order to recover the privacy-sensitive image, we embed the start and end positions of the selected private region into the camouflage image to generate a new camouflage image, called protected image. The above described steps are illustrated as Figure 1.1. In the recovery process, we use the pre-selected background image and the protected image to recover the privacy-sensitive image losslessly. The details of the processes will be described in Chapter 3.

1.3.3 Brief description of proposed method for protection of private motion activities in surveillance videos using a reversible prediction-based mapping

Based on the method of protecting selected private regions described previously,

7

we propose in this study a second method for protection of private motion activities in surveillance videos. The process of this method is illustrated in Figure 1.2.

Figure 1.1 Proposed process of protecting selected private regions.

At first, we detect motion events and decide the region of each motion event, called protected region. It is pointed out that the color difference between the pre-selected background image and the privacy-sensitive image results in an unnatural camouflage image and visible stitching seams. In order to overcome this drawback, we change the color of the pre-selected background image to suit the privacy-sensitive image. In addition, we modify the appearance of the background image in the protected region by a seamless blending technique proposed in Poisson image editing [30] in order to eliminate the visible stitching seams. After these steps, we use the resulting background image and the privacy-sensitive image to create a camouflage image by the previously-mentioned prediction-based mapping function. At last, we embed into the camouflage image the start and end positions of the protected region

8

and the parameters which are used to compute the modified color of the pre-selected background image. The details of the color transfer and the seamless blending techniques will be described in Chapter 4.

Figure 1.2 Proposed process of protecting private motion activities.

1.3.4 Brief description of proposed method for reversible image steganography by a prediction-based mapping

With the advance of computer technologies and the proliferation of the Internet, the intellectual property of various information becomes more and more important.

Plenty of methods have been proposed for copyright protection. Stealing the idea of creation or plagiarizing others’ works is also a kind of copyright violation. However, such behaviors are difficult to define and prevent. Image steganography is a feasible

9

strategy to avoid such events. A reversible image steganography method is proposed in this study for protection of a secret image. Based on the above-mentioned prediction-based mapping function, the method may be used to conceal a secret image from unauthorized people. The idea of the method is to disguise the secret image as a target image which is provided by the user. Different from the methods described in Sections 1.3.3 and 1.3.4, this method can be used to recover the secret image with the camouflage image only by embedding a sequence of recovery information.

Figure 1.3 Image steganography process by using the prediction-based mapping.

At first, we modify pixel values of the target image according to a pre-defined quantization table. Next, we use the resulting target image and the secret image to produce a camouflage image by the previously-mentioned prediction-based mapping function. Then, we divide the resulting camouflage image into 2×2 blocks and

10

categorize them into four groups according to the difference between the mean of the pixel values of the block in target image and that in the camouflage image. For each block, we check its category and record the corresponding index in a sequence, called a recovery sequence. For each pixel in the blocks which belongs to the fourth category, we recalculate the pixel value in the target image and the pixel value in the camouflage image. After these steps, a camouflage image is created. Finally, the recovery sequence is embedded into the resulting camouflage image by a lossless LSB-modification scheme [22]. With the recovery sequence, the modified target image and secret image can be retrieved losslessly later. The details of these processes will be given in Chapter 5.