Basic Principles Of Color Coding

Published Date: 02 Nov 2017

Video watermarking is an emerging discipline consisting in inserting a robust and imperceptible mark in the medium [1]. A video is a sequence of still images, which may suggest that the adaptation of marking techniques of still images is enough to solve the problem of the labeling video. However, it is not so, because video creates new constraints (such as real time), and opportunities different from those of the still image. Video watermarking is, rather, a subject of growing study for the community of digital watermarking. However, the techniques to the video appear only recently. In a first step, we will present an overview of the basic principles of video compression. Then, we will briefly explain the principles of compression formats that are most commonly used. Finally, we will examine the basics of digital watermarking to continue the presentation marking techniques being developed by researchers.

Compression Format

In recent years, the development of compression and video processing equipment allowed the digital age to flourish, and gradually to replace the analogue age. The purpose of developing compression formats and video treatments in general, is to optimize the content in order to reduce the storage space while maintaining excellent quality. This quality has motivated the expansion of research in the evaluation of video quality. Actually, it is necessary to analyze a compression system, in order to develop an effective watermarking algorithm that is imperceptible and robust to various attacks, while having sufficient capacity that enables it to insert the amount of information necessary for the application sight. But this analysis is to be preceded by a brief description of the basic principles of video compression. The purpose of such a compression system is to eliminate duplication of a spatio-temporal medium in order to reduce the video size. In the analog, these redundancies are exploited via color coding that is based on the vision and interlacing techniques. The color coding is to determine an area which approximates at best the characteristics of human vision. Many video standards such as the PAL, NTSC, and MPEG introduce a model of the human visual system to process color. These standards take into account the effect of nonlinear perception of luminance, the organization of the color channels, and the laws of visual acuity with respect to chrominance.

Basic Principles of Color Coding

Opposite color theory states that the human visual system between its decorrelated input signals black-white, red-green and blue-yellow, which are processed in separate channels. Furthermore, visual acuity for chrominance is less sensitive in luminance. Today, the old theories that limit the human vision simply to the perception of the three primary colors Red, Green and Blue (RGB), are rarely used for encoding. Instead, we commonly use color systems where the signals correspond to the differences, which are close to the model proposed by Hering’s color opponency in 1875 [2]. In video, the space resulting from these considerations is often YUV (or Y CB CR), where Y denotes the luminance, U (or CB) refers to the difference between the blue primary color and luminance, and V (or CR) denotes the difference between the primary color red and the luminance. The low acuity color allows a slight reduction of signal "color difference." For a video, it is common to use a subsampling for color coding, the commonly used terms are:

4: 4: 4 corresponds to the absence of subsampling;

4: 2: 2 corresponds to a chrominance subsampling by a factor 2 horizontally;

4: 2: 0 corresponds to a subsampling of the chrominance by a factor 2, both horizontally and vertically. This format is the one which comes as the closest visual acuity of color, when one wants to consider only one subsampling operation of colors;

4: 1: 1 corresponds to a horizontal subsampling of the chrominance by a factor 4.

Video Compression

A video takes up a lot of space. For instance, uncompressed footage from a camcorder takes up about 17MB per second of video. Because it necessitates so much space, a video must be compressed before it is put on the web. "Compressed" just means that the information is packed into a smaller space. There are two kinds of compression: lossy and lossless.

Lossy compression means that the compressed file has less data in it than the original file. This type of compression is called so because in some cases it leads to files of lower quality, because information has been "lost," hence the name. Moreover, you can lose a relatively large amount of data before you start to notice the difference. Lossy compression, usually, causes the quality loss by producing comparatively smaller files. For example, DVDs are compressed using the MPEG-2 format, which can make files 15 to 30 times smaller, but we still tend to perceive DVDs as having high-quality picture.

Whereas, lossless compression is exactly what it sounds like. It is a kind of compression where none of the information is lost. The problem with this compression strategy is that files often end up being the same size as they were before compression. This may seem pointless, as reducing the file size is the primary goal of compression.

One of the methods of video coding is to exploit the redundancy between successive images. Thus, rather than coding frame, this method encodes the difference between successive images and performs motion estimation.

The majority of video compression codecs are based on transforms such as the DCT (Discrete Cosine Transform), the DWT (Discrete Wavelet Transform), or also the fractal transformation. There are many video codecs among which are: MPEG1, MPEG2, MPEG4, and MPEG4-AVC (H.264) encoders of the future (wavelet coders). We will now give some details about MPEG2 and MPEG4 codecs.

For MPEG2 compression, motion estimation is performed using an algorithm BMA (Block Matching Algorithm). The algorithm in the standard MPEG2 is based on two principles: motion compensation and Discrete Cosine Transform. More specifically, it speaks especially of:

Temporal prediction, which is intended to exploit temporal redundancy between different images of a video sequence;

Discrete Cosine Transform (DCT), which performs a decomposition of the video signal in the frequency domain. In fact, this transformation allows exploiting spatial redundancy of an image;

Quantization, which allows the reduction of the information that is to be transmitted;

Variable length coding (e.g. Huffman coding), which exploits the statistics of the signal (a rare event encode a high number of bits, and conversely an event with a probability of occurrence to encode a large number of low bits).

Regarding MPEG4 compression, one of the main objectives of this standard is to make it interactive following the wishes of the user of the objects in scenes. This interaction is limited to the some scenarios. The design of this standard is based on the objects that represent the audio and video content. These objects, from several different scenes can be recombined to form more complex objects. This standard organizes objects in a hierarchical manner: the lowest level consists of objects called "primitives" such as still images, audio and video objects which can be used to represent 2D or 3D scenes. They can, through appropriate transformations, be placed anywhere in any video scene. The procedure will be explained in what follows as we will present the general principles of digital watermarking, as well as key algorithms of video watermarking.

Constraints of watermarking

Design of a watermarking algorithm is to find the best compromise between three constraints namely the capacity, imperceptibility and robustness.

Capacity

It is the quantity of information that can hide in the medium. It seems obvious that the greater the capacity, the brand will be more perceptible and more robust decrease (in the case where you want to find exactly the mark).

Imperceptibility

This is the impact that may have the mark on the medium. The marking will be stronger, will be visible. Ensure the invisibility of the signature returns to meet the visual quality of the media. No degradation must be charged on the original object. It is for this reason that many algorithms marking existing tend to make the insertion areas of interest less sensitive to the human eye (contour, textured area ...).

Robustness

It is the capacity of a watermarking algorithm to resist external attacks. For video, it could be simple attacks such as changing the compression format, change flow rate or any other conventional treatment. The detection should be possible regardless of the attack applied; this constraint defines the robustness of a watermarking scheme by distinguishing three types of digital watermarking:

Fragile: not resistant to any attack

Semi-fragile: not resistant to some types of attacks

Robust: resistant to all types of attacks (theoretically)

Robustness

Capacity

Imperceptibility

Fig. . Problematic in digital watermarking

Design a watermarking algorithm is to find the best compromise between these three principles depending on the application. Fig.1 provides a good illustration of this problem. Thus a watermarking scheme that is robust to a wide variety of treatments will not have a very large capacity (the watermarking techniques for protecting the rights of authors often use 64bits). In contrast, a method for hiding thousands of bits in a standard image size cannot be very robust [3].

Innocent and malicious attacks

As we saw above, one of the highlights an effective watermark lies in its robustness. Nevertheless, some basic transformations can erase marking, or at least potentially alter. All these changes, voluntary or involuntary, having a direct influence on digital watermarking, are referred attacks. Attacks can be treatments to be scrambling to either remove the protection mark in the video. Fig.2 summarizes the different types of attacks. They can also be classified based on their use [4]. In this case, there are two main families:

Attacks benevolent: These are treatments that were not initially designed to prevent detection of the brand. This may be due to degradation with MPEG2, for filtering (noise reduction), with a change in resolution. Another treatment is commonly used in the video analog / digital conversion and inversely. Finally, some geometric distortions can be used: scaling, windowing, rotation, etc...

Malicious attacks: Many attacks initially developed for still images can be easily adapted to the video with the aim to find the marking. Thereby alter the watermark either by erasing the signature or by the desynchronization of the signature, or by using the weaknesses of the marking algorithm.

Attacks

Direct attacks on the mark

Geometric attacks

Attacks protocols

Cryptographic attacks

Fig. . Different types of attacks against the watermarking

Different methods of video watermarking

The watermarking of video can be devided in 4 types:

Watermarking schemes derived from still images

Watermarking method Spatio-temporal domain

Watermarking method in the temporal domain

Technical of Video watermarking from diagram image

The principle of these methods is to apply a watermarking algorithm for still images each image component video independently.

We include in this paper some techniques that can adapt to the video are:

I.J.Cox et al. [5] have presented the approach based on spread spectrum. They offered algorithm for watermarking images, and the principles for watermarking which can be generalized to audio, video and multimedia. They proposed to insert a watermark into the spectral components of the data using techniques analogous to spread spectrum communications, hiding a narrow band signal in a wideband channel that is the data. They claim that the insertion of a watermark under this regime makes the robust watermark to signal processing operations (such as lossy compression, filtering, digital-to-analog and analog-to-digital requantization, etc.), and common geometric transformations (such as cropping, scaling, translation and rotation) provided that the original image is available and can be successfully registered against the transformed image watermark. In these cases, the watermark detector unambiguously identifies the owner.

C.T.Hsu & J.L. Wu have presented in [6] the method of watermarking a compressed video, it modifies the medium frequencies DCT coefficients based on the neighboring blocks (spatially for intra-frame and temporally for P and B frame). It forces the value of the coefficients to be lower or higher than the coefficients of the neighboring blocks, according to the value of sample of the mark that we want to insert. The extraction of the mark requires the use of video unlabelled

J. Dittman et al. [7] presented two methods of watermarking:

The first is adapted from Fridrich [8], which is to insert a pattern created by a pseudo-random generator, on a image. The energy the pattern is concentrated in the low frequencies, which ensures a greater robustness to compression, but makes the brand more visible. The main disadvantage of this system is that it requires the original image. The information contained in the brand is very poor and the system allows only whether the image is sharp or not. Both problems were solved by J. Dittman et al., Then to be adapted to video

The second method has been presented is the integration algorithm in the field DCT, based on the algorithm of E. Koch & J. Zhao [19]. The marking is done on the luminance. The luminance information is first transformed in the DCT domain (on blocks 8*8), 3 DCT coefficients are then selected at random, then, according to the information bit to be inserted, a predefined relationship is imposed on these 3 coefficients. The extraction takes place in the same way, 3three coefficients are selected and then depending on their configuration, a '1' or a '0' is extracted. The advantage of this method is the ease of integration into a compression scheme MPEGx type. But the blocks are modified independently of their content, which can lead to the appearance of annoying artifact. In addition, the algorithm is not robust to scale changes and rotations.

Video Watermarking method in the space-time domain

Given that the video signal is considered as a 3D signal, different algorithms can be applied taking into account the third dimension (time). We propose the method of wavelet decompression as suggested [9]: this scheme allows to hide only one bit, but has the advantage of generating a watermark that depends both on the copyright protection into digital video and the content of it. The advantages of this technique are robust to temporal changes such as removing or inserting images, medium images, temporal interpolation..... Its disadvantages are its low capacity since it can extract only one bit of information and the need of the original sequence in original during the operation of detection [26].

The authors of [23] propose a robust digital video watermarking procedure to embed the watermark image into digital video frames by modifying its variable-temporal length 3-D DCT coefficients. First, we generate a binary watermark image in spatial domain. Then a discrete cosine transform is applied on spatial and temporal domain to a group of blocks in successive frames obtained by the using the scene detection algorithm, to obtain variable-temporal length 3-D DCT coefficients. The method adopted embeds the watermark (binary logo image) in an uncompressed video sequence by modifying the values of the mid-range coefficients of 3-D DCT block (as the sensitivity of human eye in this frequency range is minimum). This preserves the perceptual quality of video sequences. Since each block is composed of space and time details of the video, the watermark is then spread in both spatial and temporal domain.

In [24], Radu Ovidiu PREDA and Nicolae VIZIREANU presented a new approach to spatial domain watermarking for uncompressed videos, robust against most spatial, temporal and compression attacks. The watermark detection process does not require the original video or the original watermark. The novel video watermarking technique based on the quantization of the luminance values of pixel blocks. The algorithm uses error correction codes to protect the inserted watermark and temporal redundancy to embed the same watermark in different frames of the video.

Video Watermarking method in the temporal domain

These methods hide the watermark in the time domain, changing only very low spatial frequencies (average image)[27]. Consequently, the watermark will be robust to geometric distortions of the image, but will be sensitive to temporal modifications (eg interpolation). The value of the mark will be constant over a given image, but will vary from one image to another. This can obviously lead to the appearance of artifacts, blinkings. To avoid this, first mark is weighted by a visual mask spatial and / or temporal, and secondly we use only low spatial frequencies to slow variation of the average luminance. The detection is a measure of the correlation between the subsequent of the averages of the image and the watermark that we will expect to find. There may be mentioned techniques [14, 16] which resume this general principle while improving it and [15] which propose an alternative method does not rely on the average luminance, but on isolated pixels.

We can cite the schemes basing on image mosaic, [28] proposes to generate a mosaic image from different of video frame and insert the mark on the entire mosaic background. Koubaa et al. [29] have created a brand adaptive mosaic using spread spectrum technique and applied a Sobel filter to select the regions of interest.

M.Kouba [29] presented a new approach to watermarking approach based on the detection of moving shadow areas. Indeed, the shadow is a source of information on the scene because the human eye is less sensitive to changes at the pictures more or less dark. The method of watermarking is then adapted to these conditions in order to increase its energy level images of the scene that represent low light.

A. Kerbiche et al.[30] proposed two new approaches to video watermarking robust and invisible based on regions of interest. The first inserts in the mosaic signature while the latter inserts in moving objects.

Evaluation of Watermarking algorithm

To evaluate a watermarking algorithm, some factors that influence the quality of the image or video and attacks must be identified. In fact a measure of quality must consider the visual aspect that remains a subjective way to determine the quality of a signal handling. Several studies ([21, 13]) have focused on the definition of new measures that are as objective as possible and take into account the Human Visual System. All these proposed metrics are derived from a metric that is used very often PSNR (Peak Signal to Noise Ratio). A derivative of this metric which considers the signature inserted in the image or video is one of wPSNR. We present below the definitions of these two metrics.

Evaluation of the video quality : PSNR and wPSNR

The PSNR

This type of measure returns a numeric value indicating the distortion caused. The most used is the mean square error (Mean Square Error, MSE) computed between the pixels of the two images to be compared:

MSN : Mean Square Error Module

The MSE is the mean square error between the original and the watermarked image. At the end of to assess the influence of brand image on this measure assesses the influence of brand image. It is defined as follows:

()

Where P: the set of N pixels of the image, and : the grayscale images to compare.

(PSNR): Peak Signal to Noise Ratio

The PSNR is an engineering term to quantify the performance of the encoder by measuring the quality of reconstruction of the compressed image with respect to the original image. Typical values for the PSNR in lossy image and video compression are between 30 and 50 dB, where higher is better. When the two images are identical, the MSE will be equal to zero, resulting in an infinite PSNR.

()

Where denotes the maximum luminance possible.

wPSNR : (Weighted Peak Signal to Noise Ratio)

The wPSNR is a variation of PSNR, which takes into account the neighborhood of each pixel. This metric is based on the sensitivity of the human eye to changes in textures and homogeneities regions. Value wPSNR increases when the variance is large and decreases in the opposite case. A new definition of the mean squared error (MSE) is then:

()

Then

()

Signature detection tools

Correlation factor

The measure of "degree of reliability" detected data, returns to "calculation of distances" between the inserted and detected data. This measure is carried out using the correlation. The correlation of two signals consists to measure them dependence.

The correlation between two images X and Y is to calculate the correlation between two matrices X and Y of the same size by using the following formula:

()

Accuracy Ratio

AR: calculated between the original signature and that extracted from the watermarked image [31].

()

Where,

NCB: Number of Correct Bits;

TNB: Total Number of Bits.

Conclusion

Compared to image watermarking algorithms, few techniques exist in the literature for explicitly embedding watermark data into digital video signals. However, video is nothing more than a sequence of still images, so intuitively image watermarking approaches may be easily extended into the temporal dimension.

With respect to computational complexity, it was found that the temporal multiresolution algorithm proved twice as expensive as the most complex frame-by-frame algorithm, but the per-frame cost of the former does not increase with the length of the video signal. Resilience to signal processing was only measured using operations unique to video: frame averaging, reordering, downsampling, and lossy (MPEG) compression. For each of these experiments, the temporal multiresolution algorithm proved far more resilient to signal processing than the frame-by-frame techniques.