Presentation is loading. Please wait.

Presentation is loading. Please wait.

An Efficient MSB Prediction-based Method for High-capacity Reversible Data Hiding in Encrypted Images 基于有效MSB预测的加密图像大容量可逆数据隐藏方法。 本文目的: 做到既有较高的藏量(1bpp),

Similar presentations


Presentation on theme: "An Efficient MSB Prediction-based Method for High-capacity Reversible Data Hiding in Encrypted Images 基于有效MSB预测的加密图像大容量可逆数据隐藏方法。 本文目的: 做到既有较高的藏量(1bpp),"— Presentation transcript:

1 An Efficient MSB Prediction-based Method for High-capacity Reversible Data Hiding in Encrypted Images 基于有效MSB预测的加密图像大容量可逆数据隐藏方法。 本文目的: 做到既有较高的藏量(1bpp), 同时具有可逆恢复原始图像(即较高的PSNR=Inf) 作者:来自法国 Source: IEEE Transaction on Information Forensics and Security, Vol. 13, No. 7, pp , July 2018 Author: Pauline Puteaux and William Puech Speaker: Guodong Su (Renkai Yang ) Date: 2018/11/04

2 Outline Introduction Related work Proposed scheme Experiments
Conclusions Comments 大纲

3 Introduction(1/3) Medical image secret
现在大家都无意的将个人资讯通过网络直接赤裸裸的共享给他人或者存储于第三方平台,而这些图像往往有可能被盗用于不同场合, 或者被用于分析你个人的日常行为等等。这或许将导致你的个人隐私遭到泄露。 类似于我在网上可以找到这张医学图片一样,该医学图像内容就都被公众所知道,没有隐私可言,甚至有恶意行为产生。 与此同时,第三方可能需要在你的图像上,logo上自己的一些商标或者secret,会造成原始图片的破坏,特别是secret较大时,图像质量下降明显。 因此,要寻求一个能解决上述问题的方案。 secret

4 Introduction(2/3) 显然,有必要对个人资讯(假设是original image)先做加密(key1),在把加密图像共享给他人或者存储于第三方。 而第三方把自己secret加密(key2)后,藏入加密图像中。这样做使得,第三方看不到你的原始资讯,而其他人也无法看到你的资讯。 同时,第三方可以加上自己的secret(如版权声明,数字签字等),实现秘密隐藏。 当然,我们希望这些行为都是可逆的,就是所有步骤都可以恢复为最初的样子。也就是说: 如果你有key1,你就可以恢复出original image; 如果你有key2,你就可以恢复出secret; 如果你有key1和key2,你可以恢复出secret and original image.

5 Introduction(3/3) It is difficult to hide using classic scheme
Payload PSNR 我们都知道,若要藏的secret的数据量越大(payload),对加密图像的破坏就越大,导致图像品质下降(PSNR)。对于加密图像也是如此。 加密后的图像的像素分布具有很强的随机性,相邻像素之间的关系性很低(原始图像的具有邻域特性)等,使得隐藏大容量的secret与实现RDH变得更加艰难。常规的可逆信息隐藏方法手段(直方图等)很难被使用到加密图像中。所以,我们要尽可能的在payload和PSNR之间找到一个平衡点,使得他们都尽可能最大化。这也是本文的目的。 Trade-off

6 Related work(1/2) --Two possible RDHEI schemes 藏量不高
下面介绍两种常用的RDHEI模型。VRAE和RRBE。 VRAE:图像先加密,然后在加密图像上找到空间(或方法)去隐藏secret RRBE: 图像加密前,先预留一些可用于隐藏的空间(要尽量保证可逆),然后在加密图像。Secret藏在预留的空间内。 Vacating room after encryption (VRAE)

7 Related work(2/2) --Two possible RDHEI schemes
使得一些現有的圖像可逆信息隱藏方法可以直接使用。 --Two possible RDHEI schemes RRBE: 图像加密前,先预留一些可用于隐藏的空间(要尽量保证可逆),然后在加密图像。Secret藏在预留的空间内。 Reserving room before encryption (RRBE)

8 Proposed scheme(1/14) ----Overview of generally encoding method
Ke: Encrypted key Kw: Data hiding key 为此,本文提出了基于MSB的high payload and high quality image的方法。具体是: 先对original image进行MSB误差预测,后面详细介绍。 然后根据Ke对图像加密。 secret隐藏在加密图像的MSB中。 Error location binary map is not necessary for our proposed method.

9 Proposed scheme(2/14) - Prediction Error Detection
predictor Computed from the previous scanned neighbors pred (i , j ) is treated as a predictor, it can be average value of the previous scanned neighbors, or one of the previous scanned neighbors, and so on. 预测基本原理:预测的像素值pred与真实像素值p的差距,一般一定小于 预测的像素值pred与inverser的像素值inv的差距; 利用的还是像素之间的邻域相似度高的特点 (预测是对原始图像预测,不是对加密图像) Principle: prediction value is closed to original pixel value than the inverse pixel value

10 Proposed scheme(3/14) - Prediction Error Detection
... 78 65 154 50 55 ... 78 64 154 50 55 For binary map: 1:表示当前pixel有预测误差,需要修正 0:表示当前pixel没有预测误差,无需修正 Binary map ... 1

11 Proposed scheme(4/14) - Encrypted step
c,x0是混沌置乱的参数,可以产生毫无规则的混沌序列。 x0 是初始值,c是混沌映射的控制参数。 s是采用混沌技术产生的混沌序列。

12 Proposed scheme(5/14) - Data embedding
One bit of message after encrypted 嵌入到每个可用像素的MSB, 可用像素: 没有预测误差的像素(CPE-HCRDH)或者 没有被marked起来的像素点(EPE-HCRDH) Embedding into available pixel Message after encrypted: …01110… ...

13 Proposed scheme(6/14) - Data extract and Image recovery
Three possible outcomes: 1) the recipient has only the data hiding key Kw, 2) the recipient has only the encryption key Ke, 3) the recipient has both keys.

14 Proposed scheme(7/14) - Data extract and Image recovery
Have only the data hiding key Kw ... Extract data: …01110… Data hiding key Kw Message

15 Proposed scheme(8/14) - Data extract and Image recovery
Have only the encryption key Ke ... Step1. 对低7lsb XOR就可以,MSB 需要预测 Step2. 利用前面修正后的规则,恢复MSB;基本原理:预测的像素值pred与真实像素值p的差距,一般一定小于 预测的像素值pred与 inverser的像素值inv的差距; 利用的还是像素之间的邻域相似度高的特点(预测是对原始图像预测,不是对加密图像) Step 1. XORing seven LSB for each pixel Step 2. Notion: prediction value is closed to original pixel value than the inverse pixel value

16 Proposed scheme(9/14) - Data extract and Image recovery
Have both keys Both extract message and image recovery

17 Proposed scheme(10/14) CPE-HCRDH: high-capacity reversible data hiding approach with correction of prediction errors . EPE-HCRDH: high-capacity reversible data hiding approach with embedded prediction errors . 根据是否修正预测误差,本文提出了两种方法。 CPE-HCRDH:通过轻微修改像素值,从而修正预测误差。即把预测误差binary map为1的修正为0. 通过修改像素值,最终使得所有位置的预测误差binary map都是0。这种方法,通过牺牲一点图像品质,获得较高藏量。 EPE-HCRDH: 不修改像素值,但是在预测误差binary map为1的块内(8bits),不藏secret(因为藏入后无法预测)。这种方法通过牺牲藏量,获得高品质图像

18 Proposed scheme(11/14) - CPE-HCRDH Approach
High-capacity reversible data hiding approach with correction of prediction errors Prediction error Original image I Error location binary map Modify the pixel value s.t. binary map is 0

19 Proposed scheme(12/14) - CPE-HCRDH Approach
High-capacity reversible data hiding approach with correction of prediction errors ... 78 154 50 63 65 66 62 p(i,j) pred(i,j) inv(i,j) 255

20 Proposed scheme(13/14) - EPE-HCRDH Approach
High-capacity reversible data hiding approach with embedded of prediction errors In EPE-HCRDH, one prediction error need two flags by replacing the MSB of each pixel in the previous(beginning flag) and the following blocks (end flag) by 1 : Left pixel p(i,j-1) or top pixel p(i-1,j) ... 78 154 50 ... 178 154 50

21 Proposed scheme(14/14) - EPE-HCRDH Approach
High-capacity reversible data hiding approach with embedded of prediction errors In EPE-HCRDH, one prediction error need two flags by replacing the MSB of each pixel in the previous(beginning flag) and the following blocks (end flag) by 1。 以8bits为一个块(单位)。若当前块内有预测误差值为1的,就不藏secret;同时,为了能够在提取secret的时候,知道哪个块没有藏secret, 做了两个flag( )。 当你以8bits为单位扫描的时候,若出现连续( )的时候,代表从这里开始没有藏secret, 直到下一个连续出现( )。 Prediction error Original image I Error location binary map Modify and Flag the binary map Beginning flag End flag MSB in the image

22 Experiments(1/6) CPE-HCRDH b) Histogram of the estimated
prediction errors; a) Errors’ location, number of errors = 1, 242 (0.47%); d) Encrypted image Ie c) Pre-processed image I, PSNR = d B; f) Reconstructed image I, PSNR = d B, SSIM = e) Marked encrypted image Iew;

23 Experiments(2/6) EPE-HCRDH
Unmarked pixels’ location (errors and flags), number of errors = 1, 225 (0.46%); Encrypted image Ie ; Encrypted image Ie with the highlighted prediction errors Marked encrypted image Iew; Marked encrypted image Iew with an embedding rate = bpp ; Reconstructed image I, PSNR → +∞, SSIM = 1

24 Experiments(3/6) Performance measurements

25 Experiments(4/6) Payload measurements
Payload measurement for EPE-HCRDH approach, on a sample 500 images

26 Experiments(5/6) Quality evaluation of the obtained images with out two approaches Correlation: 像素与邻域的关联度,分为水平和垂直方向; Entropy: 熵越大, 分布越随机--均匀分布 X2: 描述两个变量之间的相似性,越小,说明越混乱无序; NPCR: 没有进行预测误差修正的像素比例; UACI: 预测误差修正的平均值大小,即平均每一个像素修改的像素值大小;

27 Experiments(6/6) Performance comparisons Lena Airplane
Correlation: 像素与邻域的关联度,分为水平和垂直方向; Entropy: 熵越大, 分布越随机--均匀分布 X2: 描述两个变量之间的相似性,越小,说明越混乱无序; NPCR: 没有进行预测误差修正的像素比例; UACI: 预测误差修正的平均值大小,即平均每一个像素修改的像素值大小; Lena Airplane

28 Conclusions This is one of the first methods which proposes to use MSB instead of LSB for a RDHEI Two strategy for embedding high capacity and recovering high quality image CPE-HCRDH and EPE-HCRDH

29 Thanks!


Download ppt "An Efficient MSB Prediction-based Method for High-capacity Reversible Data Hiding in Encrypted Images 基于有效MSB预测的加密图像大容量可逆数据隐藏方法。 本文目的: 做到既有较高的藏量(1bpp),"

Similar presentations


Ads by Google