Overview of Proposed Methods

In this study, we begin by reviewing some existing techniques for privacy protection in video surveillance. Then, we propose a method for protecting selected privacy-sensitive areas in the depth and color images acquired by the KINECT device.

The main idea is easy to understand, just like the use of a “sticker” to cover a privacy-sensitive area in a video surveillance image. In order to implement this idea, a technique of prediction-based mapping is adopted [25], which involves both the color and depth images.

The prediction-based mapping technique can also be applied to other applications. Specifically, we apply it to protect privacy-sensitive motion activities in image sequences or videos in this study. In this method, the parts of the motion activities in a given image sequence or video are segmented out automatically.

Moreover, we use information hiding techniques to embed the privacy-sensitive image parts into image sequences or videos. For this, admittedly it requires a huge data embedding capability to embed the privacy information by traditional information hiding techniques. Therefore, we use a reversible mapping function which allows the mapped values to be controllable in magnitudes, and synchronizes removals of corresponding areas in both color and depth images.

More specifically, the first method proposed in this study embeds, in a way of meaningful disguise, a specific privacy area in surveillance images/videos against a pre-selected background image part, using not only the color image but also the depth image in the embedding process. The results can be shown in 3D ways. And in the second method, the first method is extended to protect the privacy-sensitive motion

5

activities by detecting human activities in images at first and regarding the detected parts as mobile privacy areas.

Finally, considering the daily-increasing popularity of the 3D images which may be constructed from the use of the KINECT device and in order to reach the goal of hiding information in 3D images, we propose as well a method for 3D steganography via KINECT images utilizing 3D image processing techniques.

The detailed descriptions of the above-mentioned proposed methods will be presented in the subsequent chapters.

1.3.1 Definitions of terminologies

The definitions of some related terms used in this study are introduced as follows.

1. Privacy-sensitive image: a privacy-sensitive image is an image which includes privacy-sensitive contents and needs to be concealed.

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

3. Camouflage image: a camouflage image is an image produced by disguising a privacy-sensitive image to make it 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. Recovery sequence: a recovery sequence is a sequence which records the location of the privacy-sensitive image and the removed bits.

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

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

6

8. Motion region: a motion region is an area containing motion objects in an input video, which are obtained from a motion detection process.

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

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

11. 3D image: a 3D image is one constructed from combining the depth and color images acquired with a KINECT device.

1.3.2 Brief description of proposed methods

(A) Protection of privacy-sensitive regions in surveillance videos acquired by a KINECT device 

A method for protection of privacy-sensitive regions in surveillance videos acquired by the KINECT device is proposed in this study. This method aims to protect privacy-sensitive images by using a reversible one-to-one prediction-based mapping function proposed by Liu and Tsai [1].

First, we generate a prediction-residue image with small pixel values. Then, we map a privacy-sensitive image and a pre-selected background image together into a third image, called a camouflage image, through the use of the above-mentioned function proposed by Liu and Tsai [1]. The resulting camouflage image is similar to the selected background image and it is hard to tell their differences by human eyes.

At last, in order to recover the privacy-sensitive image, we embed the start and end positions of the selected privacy-sensitive region into the camouflage image in order to generate a new camouflage image involving both the color and depth images,

7

called the protected image. The details of these processes will be described in Chapter 3.

(B) Protection of privacy-sensitive motion activities in surveillance videos acquired by the KINECT device 

A method for protection of privacy-sensitive motion activities in surveillance videos acquired by the KINECT device is also proposed in this study. How to protect selected privacy-sensitive regions has already been studied in the last method, so for this method we propose the use of the speeded up robust features (SURFs) for detecting privacy-sensitive motion activities of specific people in given image sequences or videos.

At first, we detect privacy-sensitive motion events in image frames, and then segment out the region enclosing each motion event in the frame, called protected region. In addition, we modify the appearance of the background image in the protected region. After these steps, we integrate the resulting background image and the privacy-sensitive image together to create a camouflage image by using the previously-mentioned prediction-based mapping function.

At last, we embed the start and end positions of the protected region and the parameters into the camouflage image which involves the color and depth images.

The details of the proposed method and the employed SURF extraction technique will be described in Chapter 4.

(C) 3D Steganography via KINECT Images 

A method for 3D steganography via KINECT images is proposed in this study. In this method, a technique is proposed to recover the secret image from the camouflage

8

image only by use of the embedded recovery information.

At first, we modify the pixel values of a given 3D cover image according to certain color and coordinate tables. Next, we embed the secret image into the cover image to produce a camouflage image. The original 3D image data we want to recover should be saved, and the recovery information for this purpose is created, called the recovery sequence. Then, we transform the resulting camouflage image into color and coordinate tables.

Finally, the recovery sequence is embedded into the resulting camouflage image by a LSB-modification scheme. With this recovery sequence, the modified cover image part during the data embedding process as well as the hidden secret image can both be retrieved losslessly. The details of these processes will be given in Chapter 5.