ES2321363B2 - DIGITAL WATER MARKING SYSTEM IN PARTICULAR OF AN AGUADIGITAL MARKING SYSTEM FOR THE DETECTION AND LOCATION OF ALTERATIONS IN DIGITAL IMAGES. - Google Patents

Abstract

Digital watermarking system in particular of a digital watermarking system for detection and location of alterations in digital images.

A digital watermarking system with digital image manipulation detection for systems video. Supervision by inserting integrity information and at least a message, characterized in that the watermarking method Understand the steps of:

transform the digital image into an image digitally transformed, divide this into a plurality of blocks; obtain a plurality of projected message values for each one of the messages; insert each bit of each message at least in one of said projected message values; get values integrity projections for such integrity information; insert each part of the integrity information into those projected integrity values; update each coefficient default used to calculate the projected values of message with the projected message values inserted; update each default coefficient used to calculate the projected integrity values with projected values integrity inserts

Description

Digital watermarking system in particular of a digital watermarking system for detection and location of alterations in digital images.

The present invention relates to a system of digital water marking in particular of a marking system of digital water for the detection and location of alterations in digital images, in which digital images are obtained in a digital video monitoring system.

In an video system monitoring an element The camera is fundamental. Currently, analog cameras are replaced by digital security cameras. these, in many cases, are designed to use communications networks existing digital, which leads to a reduction in costs of installation. That factor is the cause of the rapid increase in number of that kind of digital systems.

Other common devices in the systems of Digital video monitoring are video servers. Its function Main is to digitize the analog video signal. Other usual feature is that they perform interface functions between analog cameras and communications networks digital This allows to gradually pass a video system analog to digital supervision.

The central servers are the others devices that make up, with those presented above, the set of basic elements of a video monitoring system digital. Its essential task is the configuration of the system and the general control of digital cameras and video servers. In addition, it is very common that the sequences of video obtained.

These new digital video systems supervision appear fruit of the enormous advance in the technologies of Information in recent decades. Parallel to its appearance numerous applications arise for editing still images and video. With them it is not complicated to alter an image of so that you cannot discern between an original and a false one. In addition, the number of potential manipulators has increased from huge way, because due to the Internet those tools of Edition are available to a large number of users. A consequence of the above is that, from the point of view of the authenticity, validity of still images and videos Digital is increasingly questioned.

Digital watermarking is one of the proposed solutions to solve the mentioned problem previously. It is a set of techniques used to insert information in a digital document (image, video, audio, etc.). The introduction of the information is done by modifying the original document (guest) with the main restriction that the distortion caused by marking is tolerable (depending on the application). One of its essential advantages is that the data inserted are linked to the host, hence it is not necessary no additional files as in the case of the cryptography.

For the classification of a specific technique of Digital watermarking uses several features. Two of the most important are the robustness and the need or not of the host to make possible the extraction of information. About this last feature, it is said that a technique is blind if for information extraction does not need the guest and does not blind in the opposite case.

In video supervision, a condition it is essential that the original images are not needed to to be able to extract the information, to avoid doubling the capacity of storage required As a result it follows that a Watermarking technique for video monitoring systems must be preferably blind

A watermarking technique is robust if the brand that has inserted resists alterations, being able to be casual or intentional. In the opposite case are the fragile techniques, which are those in which the brand is corrupted after the most minimal alteration. For the detection of Fragile digital content manipulations are used or semi-fragile, because they allow to demonstrate the authenticity of the content analyzing the integrity of the brand.

Currently, one of the big families of Digital watermarking techniques is the spread spectrum, another is that formed by the so-called marking techniques of digital water with lateral information in the encoder. A particular feature of digital watermarking techniques wide spectrum in blind schemes is that they suffer the guest interference itself. In contrast, the techniques with lateral information in the encoder do not suffer such interference. Since, as indicated above, the Authentication of the images of a video monitoring system is it must be blind, the most advanced schemes use techniques with lateral information in the encoder.

An example of a digital watermarking technique with lateral information on the encoder and blind extraction can be found in the article by B. Chen and GW Wornell: "Quantization Index Modulation: A Class of Provably Good Methods for Digital Watermarking and Information Embedding" , published in IEEE Transactions on Information Theory, Vol. 47, No. 4, May 2001. This document shows the possibility of using quantifiers to insert information into the host. Basically, the proposed idea is to have a set of quantifiers of which one is selected depending on the message you want to insert. Obtaining that set of quantifiers is not trivial. The authors propose a practical procedure to obtain them in an efficient and structured way. The reconstruction points are displaced from a prototype quantifier, with the effect of obtaining a different quantifier; This technique is called Dither modulation. On the other hand, it is shown how it is possible to increase the robustness by lowering the transmission rate. One of the exposed processes consists in projecting the values of the host before inserting the mark, in this way the noise that is orthogonal to the vector on which it is projected will not influence the communication. This document presents the technique called DC distortion compensation ( Distortion-Compensated ); through it you can control the difference between the marked document and the original document. As a result, there is another value with which a valid solution can be reached between the strength of the watermark and imperceptibility.

Another approach to projection in marking digital water can be found in Fernando's article Pérez-González, Félix Balado, and Juan R. Hernández: "Performance analysis of existing and new methods for data hiding with known-host information in additive channels ", published in IEEE Transactions on Signal Processing, 51 (4): 960-980, April 2003. Special Issue on Signal Processing for Data Hiding in Digital Media & Secure Content Delivery In this document the authors give a more insight wide projection of host values, as they reach a compromise solution between insertion techniques with lateral information in the encoder and spectrum techniques widened

Another practical implementation of the techniques based on quantifiers is in the document by Joachim J. Eggers, Robert Bauml, Tomas Tzschoppe and Bernd Girod: "Scalar Costa Scheme for Information Embedding ", published in IEEE Transactions on Signal Processing, VOL. 51, NO. 4, April 2003. In This document shows a technique close to modulation Dither, but focused exclusively on quantifiers scalars

The articles presented above have as a common denominator the theoretical approach of the schemes of watermarking they propose. An example of this is modeling the Noisy communication channel as a channel with white noise additive Gaussian, while on many occasions the channel is characterized by a quantization noise; as is the case with the encoding still images in JPEG or videos in MPEG-1

There are several patents focused on the field of watermarking for image authentication. The patent US20041 31 1 84 aims to demonstrate the validity of videos to be used as irrefutable evidence before the courts. It uses Dither-QIM watermarking techniques , introducing two types of information: one is identity and the other is control. Identity information is used to identify the video sequence, and the control information is used to determine if the image was manipulated. Another basic feature in this patent is that it mentions only the MPEG standard. This standard divides the coefficients of each block by a quantification matrix, therefore there are large distortions in the information inserted at the time of compression of the group of images, as a consequence to enter the information you need to alter a large number of coefficients for each block. When marking a high number of coefficients, the difference between the original image and the image with the mark is usually quite noticeable. It is designed to be implemented in a laptop that accompanies the recording systems of patrol cars.

EP1001604 patent shows a method for entering information into images. It operates with still images encoded with the JPEG or JPEG2000 standard, and uses SCS adaptation ( Scalar Costa Scheme of Eggers et al .) To insert the information. It sets the size values of the quantification steps used to enter the information, thus reducing the versatility of the original method. In addition, it does not include any technique that allows entering information with a greater degree of robustness, such as projecting techniques.
tion.

A patent used to authenticate flows of Images is US2003172275, with the aim of guaranteeing Copyright. It classifies the images that form in synchronous and asynchronous flow. In synchronous images, enter a mark in the selected blocks pseudorandomly. In the introduction of the information you use Insertion techniques with lateral information in the encoder. Because the brand is not introduced in the whole image, it is not Possible to locate the alterations.

An idea to join the marking techniques of water and network cameras, network camera servers or Digital video servers are shown in US2004071311. It indicates a possible solution to integrate the cameras and the process of inserting the watermark from a point of view physical. There are currently numerous camera manufacturers of network, so it is more feasible to design a method that suits perfectly to existing cameras that try to design them From the beginning. The process of introducing the brand is characterized by inserting a robust to demonstrate the authenticity and other fragile to locate the alterations. Do not However, a sufficiently described method is not described in the patent complete to address the problems of compression in JPEG or in any of the MPEG standards.

From the above, the need to find a practical solution to the problem of detection and location of spatial manipulations and / or temporary in still images or image flows generated by Digital monitoring video systems. Where said solution should provide a high degree of reliability and safety, in such a way that what the images show is irrefutable. Other requirement necessary, which has not been solved yet, is the perfect adaptation of the authentication methods to the particularities of the existing digital video surveillance systems, such as the transcoding resistance from JPEG to MPEG or the adaptability to the computational limitations of the devices which integrate these systems, for example: security cameras digital

Summary of the Invention

The system object of the invention allows solve the problem described above by providing a Digital watermarking technique for detection and location of spatial and / or temporal manipulations of images for systems Video monitoring. With the additional feature of owning a High degree of reliability and safety.

The system of the invention is composed of a insertion system and other watermark extraction digital

In the insertion system of the invention, enter at least two messages and the information in the image integrity. One of the messages is a temporary identifier that it allows to associate to an image the moment in which it was obtained. Another message is a unique identifier of the origin of the image. In this way, the information extraction system from the image you can determine the device that originated the image, the moment it was taken and check the integrity with the integrity information extracted. The data obtained from origin and Temporary reference can be checked to verify your validity.

The devices of the video systems digital monitoring capable of taking images or generating them they usually use the JPEG still image standard or some MPEG family video standard. The system exposed in the invention is designed for images and videos encoded with such standards; being robust to transcoding between they.

Minimization of distortion introduced to inserting the digital watermark into an image is a restriction intrinsic of the invention, since it is required to validate the content and not modify it. This is achieved by adapting the systems to the noise introduced into the communication channel of the image coding in the standards above mentioned. According to this, the information is entered in the DCT transformed domain of image blocks, specifically in predetermined coefficients. The goal is to reduce the alteration of the predetermined coefficients and the number of coefficients needed to introduce the brand with a certain reliability In this sense, the systems of the invention have a clear advantage, since in them the generation and insertion of the brand take into account the quantification noise that will suffer the information that is entered.

From a functional point of view, the first part of the information insertion system of the invention is the selection of the blocks of the image and the coefficients of each block that will house each bit of each message and the integrity information. One way to solve it efficiently and also ensuring that communication is hidden is to select the blocks of the image with a secret key. The resolution of the detection and location of the manipulations is strongly related to the blocks in which the integrity information is entered; If this information is inserted in the whole image, the alterations can be detected and located throughout
she.

The second part is to project the values of the coefficients selected for the messages and for integrity information about projection vectors, where the size of the projected vectors, which are the vectors resulting from the projection process, it is a variable parameter of method. To increase privacy, you can generate message projection vectors with a secret key. For originate the projection vector on which the projections are projected coefficients that will contain the integrity information is used a function whose parameter is one of the messages and the key secret One of those messages is the reference or temporary stamp, so that the integrity information of an image will be depending on the moment it is obtained. By existing this temporary dependency, marked images of previous moments to fake a video stream, because the temporary stamp will not correspond to the period of time and will will detect that attempt of falsification. The size of the vectors selected projected allows to reach a balance between the brand robustness and transmission rate. In addition, the values that form each of the projection vectors, on which project the coefficients that will house each bit of the messages or each part of the integrity information, can be weighted based on the values of the quantification table used for encode the coefficients in the JPEG standard or some of the MPEG standards, or any other table obtained from perceptual considerations.

The third part in the insertion is the quantification of the values of the projected vectors with a selected quantifier of a set according to the value of the bit of the message or the part of the integrity information that is want to enter Each quantifier is produced by modifying a prototype and using a displacement vector according to the value that you want to introduce. To obtain this vector the key is used secret for messages, and for information on integrity of the secret key and the value of the messages from which it must be dependent.

The last part of the insertion system it consists of updating each coefficient of the blocks of the image selected, to carry messages and information integrity, with the result of the quantification of the vectors projected. When marking on the values of the projected vectors,  makes the orthogonal noise to each of the vectors of projection has no influence on communication, with the consequent increase in robustness.

The extraction system of the invention allows extract the messages and determine if the image was manipulated spatially and / or temporarily. The insertion and extraction system match in the way they select the image blocks and the projection of the coefficients, with the secret key. So, without knowing the secret key the messages cannot be obtained rebuilt

Once the vectors have been obtained with the projected values of the messages and information of integrity, the information is extracted. Them and achieved by subjecting these vectors to a synchronized quantification to from the key. Analyzing the distances to each centroid of the result of the quantification the messages are obtained reconstructed and it is decided for each block with the information of integrity the authenticity of it. In order to decide about Image integrity is necessary, to synchronize the extraction, use the secret key and the message on which it depends as synchronization element of the extraction. Normally said Message is the temporary reference. If it does not correspond the Temporary reference with integrity information will be noted Image blocks as fake. If the temporary reference and the brand correspond, it will be checked that the value of the Temporary reference is within the permitted thresholds; in case otherwise the method will indicate that there was a sequential break of a group of images.

As for the level of robustness, the introduction of the messages associated with a higher level of robustness than for Integrity information. This feature is derived from the need for the perfect decoding of the necessary messages to determine the integrity of the image. Without those messages, like it has been said, it is impossible to synchronize and therefore take a correct decision about the authenticity of the image blocks. On the other hand, integrity information must be very sensitive to any alteration other than the distortion of the JPEG or MPEG encoding; hence it has a low robustness.

Another object of the invention is the high reliability of the exposed system. This is achieved by achieving that the generation of integrity information for a specific image in a specific moment and a specific device, without knowing the Secret key, be a computationally unapproachable problem. In addition, determine the secret key used to insert the information analyzing still images or videos marked results also computationally difficult, which provides a high degree of security

Another object of the invention is the possibility of implement the insertion method inside the chambers of network that make up a digital monitoring video system, where those network cameras have very high restrictions on As for the available computing resources. To undertake it, the proposed system is designed to minimize the number of calculations and access to memory. For example, dial in the transformed domain downloads to the CPU numerous cycles of instruction of the calculation of the transform. Another example is the use of the least number of coefficients to insert the information minimizing the number of values to perform the calculation, as well as a perceptual improvement.

Another object of the system exposed in the invention is to allow the maximum possible versatility, being able to configure the system so that a solution can be reached of compromise between speed, distortion, data volume entered and / or error rate. One of the possible parameters configurable is the number of coefficients associated with each bit of message information or each part of the information of integrity. The relationship between the length between the vector originating and the projected vector is another possibility. He distortion control factor is the value that sharpest allows to balance the method between distortion and probability of detecting an error, two opposing characteristics in the application. Another possibility is to use channel encoders in the posts; in this way its robustness is increased by lowering the rate of error.

Other advantages and characteristics of the method they will be apparent in the figures presented in conjunction with the Description that comes next.

         \ vskip1.000000 \ baselineskip
      

Brief description of the drawings

To complement the description that is being performing and in order to help a better understanding of the characteristics of the invention, according to a preferred example of practical realization of the same, a set of drawings where illustrative and non-limiting, has been represented the following:

Fig. 1 schematically shows the structure general of the system comprising the invention.

Fig. 2 shows in a diagram the modules in those that the insertion system is divided.

Fig. 3 shows in a diagram the modules in which is the extraction system divided.

Fig. 4 represents a flow chart of the process of entering a message with the IP address in a image.

Fig. 5 represents a flow chart of the process of entering a message with a temporary reference in an image

Fig. 6 represents a flow chart of the process of entering integrity information into a image.

Fig. 7 represents a flow chart of the process of extracting a message with an IP address from a image.

Fig. 8 represents a flow chart of the process of extracting a message with a temporary reference of an image

Fig. 9 represents a flow chart of the process of checking integrity of an image.

Fig. 10 shows the division of an image into blocks and in macroblocks, it also represents the arrangement of coefficients according to the zigzag order.

Fig. 11 represents the selection and sorting of the coefficients of the blocks that form a macroblock of integrity.

Fig. 12 represents a diagram of the process of Digital watermarking insert with side information in the encoder

Fig. 13 schematically represents a internal diagram of a digital network camera implementing the insertion methods proposed in the present invention.

Fig. 14 schematically represents a computer system that executes the methods with detection of handling proposed in the present invention, reflecting the interconnection between the devices of a digital video system supervision.

Preferred Embodiment of the Invention

Fig. 1 schematically shows the elements individual comprising the selected preferred embodiment for the present invention. These elements are where you submit to a sequence of images, obtained by a video system digital supervision, to digital watermarking techniques that allow to detect and locate spatial alterations and / or Temporary suffered.

After a digital network camera gets a digital image, it is processed in an encoder 100 where the watermarking insertion system is implemented digital proposed in the present invention. The marked image can be saved as part of a video with any of the standards MPEG or as an individual image encoded according to the standard JPEG, in both cases it is stored in a storage unit 300. If you choose to store a sequence of individual images marked forming an MPEG video, they are encoded in an encoder MPEG 400

Checking the integrity of the images marked begins by retrieving, from the storage unit 300, images archived individually or in the form of video MPEG If the image stream is encoded with any of the MPEG standards the image information is extracted in a MPEG 500 decoder. In a decoder 200, which implements the digital watermarking method with tamper detection proposed in the invention, is where it is processed in order to detect and locate the alterations of the information obtained from MPEG 500 decoder or individual images stored in storage unit 300.

The data generated by decoder 200 is relative to the origin of the marked image, the instant it was taken, to the alterations detected, etc. In addition, you can configure to generate a result image by pointing out the Modifications found.

Still Image and Video Standards

In the systems of insertion and extraction of watermarks the coding standards contemplated are the JPEG for still image and MPEG family standards for the video.

One of the reasons to use in the still image coding of this practical embodiment the JPEG standard is the good ratio between size reduction achieved and the distortion introduced. Another reason is its use Majority in current digital monitoring video systems,  allowing the present invention to adapt perfectly to they.

Among the still images that make up a sequence the existing redundancy is high. This factor can be explode to achieve a notable reduction in size, achieving Make your storage more efficient. In the proposed systems in the present invention, for the coding of a sequence of Images are used by a member of the MPEG family of standards. This is due to the numerous coincident points between the MPEG standards and the JPEG standard, allowing methods proposed easily adapt to both.

The JPEG standard uses the YCC as the three-component color space. The components are luminance (Y) and color (C and C). Each component of the image is divided into non-overlapping blocks of 8 x 8 samples. JPEG uses a change of coordinates in order to concentrate most of the energy in a smaller number of dimensions than the original representation (RGB), which is exploited to reduce the size of the image. Specifically, JPEG uses the discrete cosine transform (DCT) in each block of the image of the Y, C or C components. Each transformed block of the image is subjected to the quantification of the 64 coefficients that form it, dividing them by elements which constitute a JPEG quantification matrix of dimensions 8 x8, obtaining as a result quantized coefficients c quant, j, as shown

c_ {quant, j} = round (c_ {j} / q_ {j}),

where in the previous expression round () denotes a function that returns the nearest integer value. To the quantized coefficients c quant, j of the blocks B i, a lossless entropic coding is applied to generate the JPEG file with the image, where the coding is specially designed to efficiently store those c quant , j} whose value is zero.

To recover the reconstructed transformed blocks \ hat {\ mathit {B}} {i} that form the image, the inverse process to quantification must be performed. Each quantized coefficient c quant, j pertaining to each \ hat {\ mathit {B}} {i} must be multiplied by its respective element of, that is,

\ hat {c} = c_ {quant, j} \ cdot q_ {j}.

In the way that a code is encoded and decoded image in the JPEG standard shows that the larger the JPEG quantization matrix values distortion introduced will be greater, although the compression will also be greater. The Image energy is concentrated in the lower frequencies of the transformed blocks of the image and, on the other hand, the system  Human visual is more sensitive to those frequencies; therefore in the generation of these characteristics should be taken into account to achieve a good JPEG encoding by assigning the sizes of quantification steps lower than the most frequencies low. Many times, the result of applying according to those considerations is that numerous of the highest frequencies are zero and, as mentioned earlier, a considerable size reduction

Fig. 10 shows the division of the images 1003 in the JPEG standard forming 8 × 8 pixel blocks 1001. In addition, the zigzag ordination of the coefficients that form the blocks, represented by an arrow starting at the coefficient at position (1,1) (coefficient of continued) and ends in the coefficient (8.8).

In the present description of the embodiment Preferred is that JPEG images that form a stream of images obtained by a video monitoring system will Compress with the MPEG-1 standard. A sequence of MPEG-1 video comprises several types of images, being the types I, B and P. Attending to this characteristic, the images that will contain the brand are the encoded type I and P, leaving those coded as type B without a mark so they can Withstand greater compression.

When a sequence of individual still images is encoded as an MPEG-1 stream an encoding pattern is used. For illustrative purposes we will consider the IBBBBPBBBB pattern. This pattern indicates that the first image is encoded as a type I image, the next four type B, then as type P and the last four type B. The proposed watermarking methods do not use the images marked coded as B. If If you want to increase the degree of compression of a video of marked images, you must increase the proportion of images encoded as type B. As a result, the number of images I and P in the stream decreases and, with this, the ability to detect alterations. Using the above-mentioned pattern, if a digital network camera generates images marked at a rate of 10 per second and are encoded in the MPEG 400 encoder, only alterations with a resolution of 0.2 s can be detected. According to this, it is necessary at the moment in which the user of the system configures the MPEG-1 encoding, a compromise is reached between the degree of compression and the temporal resolution of the alterations.
nes.

Information linked to images

In the present preferred embodiment of the invention the inserted mark is composed of a message of metadata, a message with a temporary reference and information of integrity

In the metadata message, arbitrary information may be inserted and extracted from the image. As an illustrative and non-limiting example, a unique identifier of the supervisory video image generator element is used. Due to the usual context in which a digital monitoring video system is located, the IP address is selected in the present practical embodiment. Note that there are numerous alternatives as a unique identifier, such as: serial number, MAC (English Media Access Control ), etc.

The information is entered in the image digital in the transformed coefficients of the blocks of luminance that form the image. The blocks in which you Enter the integrity information grouped into macroblocks. The dimensions of the macroblocks for the insertion of the Integrity information are configurable and also determine the  granularity of the detection of spatial alterations. How much the greater the macroblock the harder it is to point the points concrete of the image that were modified, although this option It has the advantage of the possibility of using a larger number of coefficients to determine its authenticity. For example, a macroblock of integrity information can be formed by 2 \ 2 blocks. A macroblock 1002 is shown in Fig. 10 used in the detection of spatial alterations.

Encoder Modules

In Fig. 2 the internal scheme of the encoder 100, consisting of an insert module of the IP address (metadata) 110, an insert module of the temporal reference 130, an information insertion module of integrity 150 and a record with the reference value temporary 170.

When an image encoded with the standard JPEG is entered in encoder 100, processed in the module insertion of IP address 110 to enter in each image the IP address of the monitoring system video device digital that obtained it. Then, the image is passed to the module insertion of temporary reference 130 to enter in the image the temporary reference (stored in register 170); with that information will be able to know the moment in which the image and also determine if the order was altered in a sequence of images (temporary alterations). Each image at the exit of the module 130 is passed to the information insertion module of integrity 150; in that module the information of integrity with which alterations can be detected Space When inserting the integrity information, use the temporary reference of register 170 so that it is not can falsify one image using another, as the information in integrity is not valid at a time other than that in which gender. Modules 110, 130 and 150 use a secret key K, of such that entering those two messages or the information of integrity without such a key is a computationally very problem complicated.

Decoder Modules

Fig. 3 shows the internal scheme of decoder 200, formed by a module for extracting the IP address 210, a module for extracting the temporary reference 230 and an integrity information extraction module 250

The module for extracting the IP address 210 retrieves the IP address inserted in the image. The module of extraction of temporary reference 230 retrieves the reference temporal that is linked to the image. The extraction module of the integrity information 250 retrieves integrity information and determines if an image was spatially manipulated, that is, if  some picture of the original image was modified. If the image was manipulated, the module for extracting integrity information 250 can generate an image indicating graphically the position of the alterations detected. The extraction modules 210, 230 and 250 use the same secret key K to extract the information than the one used to enter the information.

Internal ordering of the encoder and decoder

Although the internal schemes of the encoder 100 and of the decoder 200 of Fig. 2 and Fig. 3 show a connection cascading between modules, this feature is not restrictive It is possible to arrange the modules in parallel or even group them into a single module.

Encoder

Fig. 4, Fig. 5 and Fig. 6 represent the system flow diagram of the insert modules that they form the encoder 100, which can be seen in Fig. 2.

Flowcharts are configurable. by assigning values to certain parameters of the system. One of the system parameters is the number of Image coefficients per bit of information used in the case of the messages or the number of coefficients of each macroblock for the introduction of integrity information. In the diagrams are represented by NºCOEF_IP, NºCOEF_REF or NºCOEF_INT, in the case of the message with an IP address, the message with temporary reference and integrity information respectively. Another parameter of the introduction diagrams of the messages is the number of information bits of each message, which in the diagrams are represented by NºBITS_IP and NºBITS_REF, in the case of the IP address and the temporary reference respectively. Another possible way to configure the system is by selecting the block coefficients that will be used to insert the message with the IP address (metadata), the message with the temporary reference and information of integrity The size of the projection vectors on which the coefficient values are projected, of which get the projected values in which the information, is another configurable value, representing it by LONG_IP, LONG_REF and LONG_INT for the IP address, the reference Temporary and integrity information respectively.

Fig. 4 shows the flow chart of the IP address insertion system (metadata) 110, in which each image obtained by a video system will be processed supervision. If this is the first time the method is executed (step 111) in an image stream, in step 112 are selected pseudorandomly (depending on the secret key K) the blocks transformed by 8 x 8 coefficients into which the image, for use in inserting the IP address. At step 113 initializes the BIT variable, used to determine the number of bits of the IP address message that have been introduced in the image. In step 114 it is decided if they have already been introduced all the bits of this message in the image; yes the answer is yes the insertion part of the IP address in The image will be over. Otherwise, follow in step 116 in which the coefficient counter per bit is initialized COEF to zero. In step 117, it is determined whether the BIT bit value in the NºCOEF_IP coefficients it had assigned, if so you will proceed to step 115 which will add one to the BIT counter value and will return to step 114 described previously. If the coefficients of the image associated with that bit of the IP address is passed to step 118, where a pseudo-random projection vector is generated (depending on of the secret key K) of length LONG_IP. After getting the projection vector is continued in step 119, where the scalar product between the projection vector and the vector formed by the selected coefficients (in step 112) whose indexes are between (NºBITS_IPxBIT) + COEF and (NºBITS_IPxBIT) + COEF + LONG_IP-1, saving the projection result in RES_PROY. In step 120 you add LONG_IP to the COEF value, which will be used to determine the coefficients used in the calculation of the following projection. At step 124 the BIT bit value of the IP address is inserted in RES_PROY and in step 125 the coefficients are updated with the that the current projection was calculated with the result obtained in the previous step (124).

Fig. 5 shows the flow chart of the Temporary reference insertion module 130. As in the Fig. 4, at the beginning a digital image obtained from the video monitoring system. In step 131 it is determined whether it is the first time the method is executed for that particular flow of data. If so, proceed to step 132 which will select pseudorandomly (depending on the secret key K) the blocks of the image that will host the message with the reference temporary. In step 133 you submit the message with the reference temporary to a channel coding, whose objective is to increase its robustness producing a coded message of the reference temporary. In step 134 the used BIT variable is initialized to zero to control the number of bits of the encoded message of the Temporary reference that has been inserted in the image. In step 135 is checked if the coded message of the temporary reference It has been fully inserted into the image; for this it is checked if the value of BIT is less than the number of total bits allocated for inserting the message coded with the temporary reference NºBITS_REF. In case this condition is not met, the work of the insertion module of the temporary reference 130 there will be finish. If bits of the temporary reference are still subtracted by enter is continued in step 136, which initializes the COEF value to zero The COEF variable is used to control the number of coefficients that have been entered from the BIT bit of the message Coded temporal reference. In step 138 it is checked if the NºCOEF_REF coefficients assigned for host each bit of the temporary reference, if applicable. Yes already have been completed, continue in step 137 that updates the BIT value by adding one, indicating that the following is will proceed to insert the next bit of the coded message of The temporary reference. If you have not yet used all coefficients corresponding to each bit of the time reference, step 138 to 139 is advanced. In step 139 a vector is generated LONG_REF projection length, obtained pseudorandomly With the secret key. This projection vector will be used for project the values of the following LONG_REF coefficients. In step 140 calculates the scalar product between the vector of projection (obtained in step 139) and a vector formed by coefficients selected in step 132 whose indices are found between (NºBITS_REFxBIT) + COEF and (NºBITS_REFxBIT) + COEF + LONG_REF-1, storing the projection result in RES_PROY. In step 144 you add to the COEF value the value of LONG_REF. Then, go to step 145, in which the bit of the RES_PROY is inserted in the projected value coded message of the temporary reference whose index is BIT. In step 146 updates the coefficients that were used to compute the current projection with the value of RES_PROY, result from step 145.

Fig. 6 shows the flow chart corresponding to the module for inserting information from integrity 150. The value is passed to this system as a parameter of the temporal reference that is introduced in the image by the temporary reference insertion module 130 obtained from register 170 and the secret key K. In step 151, the system to get the introduction of information from image integrity is dependent on the moment it was taken the image that is being processed is also used in the synchronization of the secret key K to ensure that only authorized users can enter the integrity information valid. If the temporary reference is not known, the process of enter the integrity information without it in the decoder 200 of Fig. 1 detected is very complex. In step 152 you initializes the index value i, whose function is to count the row macroblock in which the insertion method is located during its execution. In this figure, the block width (8 \ x8) of the image is denoted by WIDTH and the height in blocks per HIGH, the block width of each macroblock M and the high in blocks of each macroblock N. In step 153, check if the integrity information has already been entered throughout the image, if so, the insertion process of the integrity information If the totality has not yet been computed of the image is advanced to step 154 in which it is initialized to zero the index value of the macroblock columns j. In step 156 verifies if all columns of macroblocks of the current macroblock row, row i. Yes, I know have computed follow step 155 which adds a unit to the index i to process the next row of macroblocks or finish, deciding this in step 153. If the result of the decision taken in step 156 is that macroblocks of row i remain process is continued in step 157, where a vector is created VEC_INT of length NºCOEF_INT, taking from each block, which forms the coordinate macroblock (i, j) of the image, NºCOEF_INT coefficients The order of arrangement of the coefficients of each block in said vector VEC_INT is from left-right and top-down regarding the position of the blocks in the current macroblock, as shown in Fig. 11. The order of arrangement of coefficients belonging to each block in the vector VEC_INT follow the order of the coefficients according to the zigzag order of those own coefficients in the block. In the next step, the 159, the COEF value is initialized to zero. Then, in step 160, check if the integrity information has already been inserted in the current macroblock (i, j), if yes, a unit is added to j in step 158 and continue in step 156. If decided (step 160) that there are still unmarked blocks (COEF <NºCOEF_INT) continue with step 161. In step 161, it is generated pseudorandomly based on the temporal reference and the secret key a projection vector of length LONG_INT, where the generation of this vector is a result of synchronization carried out in step 151 of Fig. 6. In step 162, compute the projection, for this the scalar product is calculated between the projection vector (obtained in step 161) and a vector formed by the coefficients of VEC_INT with indices included between COEF and COEF + LONG_INT-1, where the result of This projection is saved in RES_PROY. In step 163 the LONG_INT value to COEF value. Then, continue on step 168. It inserts the value of that part of the information of integrity in the RES_PROY value, which is the result of the projection of the image coefficients on the vector of projection. In step 169 the coefficients are updated used to calculate the current projection with the value the value RES_PROY marked, result of step 168.

Decoder

In Fig. 7, Fig. 8 and Fig. 9 the flowcharts of the processes that are performed in the modules of extraction that form the decoder 200 and that can be see in Fig. 3. The values of the parameters used in the encoder 100 must be the same as those used by the decoder 200. For example: NºCOEF_IP, WIDTH, M, NºBITS_REF, etc.

Fig. 7 shows the flow chart of the process of extracting the IP address corresponding to the module of extracting the IP address (metadata) 210. At the beginning it find a digital image from which module 210 will extract the IP adress. In the first step (211) it is checked if it is the first once the flowchart is executed in the sequence to which The image to be processed belongs. If yes, continue in step 212 in which they are selected pseudorandomly with the secret key K the blocks of the JPEG image from which the message will be extracted with the address IP. The next step is 213, in which the value is initialized BIT to zero. That value allows you to control the number of bits in the address that are extracted. It is in step 214 where it is verified if the total bits of the message with the IP address were recovered; if so, the address extraction is completed IP of the current image that is processed in this module. If I still don't know has finished with the total of the bits that form the IP address NºBITS_IP is continued in step 216, where the initialization is reset to zero COEF value The COEF value is a counter of the number of coefficients that have been used to retrieve the BIT bit value of the IP adress. The decision on whether the total has already been used of coefficients NºCOEF_IP needed to extract the value of a bit of the IP address is taken in step 218. If it is already finished it will continue with step 217, in which the value of the BIT bit of the IP address using a vector VEC_DIS of length COCO_IP / LONG_IP Nº formed by the values stored in step 226. In step 215 the value of the BIT variable is updated for continue or end with the extraction of the IP address, which decide at step 214. Step 219 is reached if the decision taken in 218 is that there are still groups of coefficients to process for be able to calculate the value of the BIT bit of the IP address. In step 219 a projection vector of length LONG_IP is generated, where that vector must be equal to that generated in 118 of Fig. 4, to this must be known the secret key value K. In step 220 you compute the projection using the scalar product between the vector projection and a vector formed by the coefficients (selected in step 212) whose indexes are between (NºBITS_IPxBIT) + COEF and (NºBITS_IPxBIT) + COEF + LONG_IP-1, finally store the value in RES_PROY. In step 221 the value is added LONG_IP to COEF. In step 226 it is stored in the COEF / LONG position IP of the VEC DIS vector the result of calculating the distance between  RES_PROY and a reference vector. This last calculation will be explained later in this description of the embodiment preferably in detail.

Fig. 8 shows the flow chart of the process of extracting the temporary reference 230. Same as previous flowcharts described, at the beginning there is a digital image from which you want to extract the temporary reference That has bound. The first step (231) of the diagram determines if it is the first time it is extracted from the image of the present flow of images, if so, proceed to step 232, and otherwise to step 233. In step 232, pseudorandomly selected with the secret key K the blocks of the encoded image that will be used to extract the temporary reference. Step 233 initializes the BIT value to zero; with that value you can know when the process of extracting the temporary reference in the image. In step 234 it is decided if all the bits of the temporary reference message in the image, continuing in step 235, or if they still remain, continuing in the step 237. Step 235 is composed of a channel decoder that gets the value of the message from the temporary reference processing the values of the VEC_DIS vector, the process being inverse to step 133 in Fig. 5. After decoding channel is reached at the end of the process described by the diagram of flow. In step 237 the zero value is entered in COEF, which is use as a counter the number of coefficients used in a determined moment to extract the corresponding metric to the bit BIT of the coded message of the temporary reference. In step 238 it is decided if all the coefficients used have already been computed to extract the metric corresponding to one bit of the message Coded temporal reference. If all have already been computed the coefficients associated with a given BIT bit are continued with the sum of a unit to the BIT value to calculate the values of RES_PROY of the next bit of the message, if it exists. In case of that all the necessary coefficients have not yet been processed to extract the information inserted in the image relative to a bit is continued in step 239. In step 239 a vector is generated of projection of length LONG_REF on which the coefficient values. In step 240 the product is calculated scale between the projection vector and a vector formed by coefficients selected to insert the temporal reference (selected in step 232) whose indexes are between (NºBITS_REFxBIT) + COEF and (NºBITS_REFxBIT) + COEF + LONG_REF-1, the value obtained is stored in RES_PROY. In step 244 the COEF value is updated by adding LONG_ REF. In step 246, the value RES_PROY is taken and its distance from a reference vector, whose result is stores in a vector VEC_DIS in the position COEF / LONG_REF + BITx (NºCOEF_REF / LONG_REF), where VEC_DIS will be used in step 235 to determine the reference message temporary, as indicated above. The way to calculate the RES_PROY distance will be addressed later in the description.

Fig. 9 shows the flow chart corresponding to the process of extracting information from integrity 250. This method is based on the reference temporary extracted in module 230 and the secret key K; without him correct value of the temporary reference could not retrieve the integrity information and the image would be flagged as false by not correspond the temporal reference that has linked to the integrity information The flowchart has as input an image from which the integrity information is to be extracted and Check its authenticity spatially. In step 251 the method is synchronized with the temporary reference and the secret key K, which will allow to correctly generate pseudorandom values necessary for the present extraction process. In step 252 you initializes to index zero the index i that will be used by the method to know in which row of macroblocks it is. In step 253 it is determined whether the extraction of integrity information has come to an end, knowing when checking if they have already been computed All rows of macroblocks. If all have not yet been computed the rows continue with step 254, in which the j value that allows you to control the number of macroblock columns of a row of macroblocks i that have been computed. In step 254 it is decided whether the integrity information has already been extracted from all the macroblock columns that make up each row. If that is thus, a unit is added to the counter i in step 255 and returned to the step 253. In case you have not yet computed all the columns are passed to step 258 in which the vector of Macroblock coefficients with coordinates (i, j) used to extract integrity information. The way to generate the vector it is the same as the one shown above in step 157 of Fig. 6. In step 260 the COEF value is initialized to zero, which allows control the number of coefficients used to extract the information for each macroblock. The decision on whether they have already computed all the coefficients of a macroblock is taken in the step 262. If so, proceed to step 261 in which the vector distances VEC_DIST and it is decided if the macroblock was manipulated; if this is the case, continue in step 259. The function from step 259 is to designate a macroblock as false; this can be do pointing to the macroblock in some visible way and / or issuing  a signal that allows a user to know that this macroblock has been altered If in step 261 it is not decided that it has been manipulated the block studied is continued, just as after the step 259, with step 257. In step 257 a unit is added to the counter j, indicating that the extraction of information from Integrity of a macroblock is over. On the other hand, if in the step 262 it is decided that the information from has not yet been recovered integrity corresponding to that macroblock is continued in step 263. In step 263 a projection vector of length is generated LONG_INT synchronized with the generator at 161 in Fig. 6. Its generation is dependent on the temporary reference and the key, hence, if the temporary reference or password is unknown secret will be determined with high probability that the macroblock is false. In step 265 the scalar product is made between the projection vector and a vector formed by coefficients of VEC_INT with indexes between COEF and COEF + LONG_INT-1, where the projection result It is stored in RES_PROY. In step 266 the COEF value is updated adding the value LONG_INT. After the current projection has been completed, continue with step 264. In that step you enter the result value of computing the distance between RES_PROY and a reference vector in the result position of the COEF / LONG_INT division of the distance vector VEC_DIST. The explanation of the distance calculation will be addressed later in the description.

Pseudorandom block selection

In the present practical embodiment, the blocks DCT transforms that house the messages of a digital image Fixed encoded according to JPEG standard are selected pseudorandomly. It takes place in pairs of steps 112 (Fig. 4) - 212 (Fig. 7) and 132 (Fig. 5) - 232 (Fig. 8) for message with the IP address and the message with the temporary reference respectively. If the blocks selected in coding and decoding are not the same the recovered data are Invalid

The selection is made by swapping the blocks that form the image using a number generator pseudorandom, in which the value of the secret key or a function of it. The transformed blocks of the image is arranged forming a vector of blocks attending to the left-right order and of up-down with respect to the position of the blocks in the image. With the pseudorandom generator, the position of each block in the block vector obtaining a vector  of messy blocks.

The coefficients of the blocks that are set they form the vector of disordered blocks for each of the messages that are going to be hidden in the image. Coefficients associated to each message are arranged forming a vector of coefficients c. The vector coefficients of the message with the IP address is denoted by and has a length of NºBITS_IPxNºCOEF_IP. The message with the temporary reference is denotes by with a length of NºBITS REFxNºCOEF REF ..

If vectors or projection matrices do not are orthogonal, it must be met that the sets of coefficients associated with each message are disjoint. If this is not fulfilled, the insertion processes of each message would be interfering each other and the information inserted could not be recovered.

The pseudorandom selection of blocks is you can perform when each image is going to be processed and not only at Beginning of a flow of images. This possibility has the drawback that the runtime of the insert or of the extraction increases so this solution is not advisable in those cases in which there are temporary restrictions.

Synchronization function of time reference and key secret

The timing of the introduction and the Extraction of integrity information is performed in the steps 151 (Fig. 6) and 251 (Fig. 9). The secret key K is used to ensure that integrity information is entered by a authorized user of the system. Synchronization with the seal temporary is implemented with the objective of creating a dependency between the process of inserting the integrity information and the Temporary reference value. As a result, it is not possible use images marked at an earlier time to falsify an image obtained at a different time without It is detected. This synchronization is achieved by initializing a generator pseudorandom integrity with a value that is a function of the Temporary reference and secret code. The generator pseudorandom is used in the part corresponding to the control of integrity in obtaining values in steps 161 (Fig. 6) and 168 (Fig. 6) in the introduction, 263 (Fig. 9) and 264 (Fig. 9) in the  extraction.

Generation of projection vectors

The pairs of steps 118 (Fig. 4) - 219 (Fig. 7) of the message with the IP address, 139 (Fig. 5) - 239 (Fig. 8) of the temporary reference message and 161 (Fig. 6) - 263 (Fig. 9) of integrity information generate projection vectors, and of length LONG_IP, LONG_REF and LONG_INT respectively. The projection vectors are obtained pseudorandomly by making them dependent on the secret key for messages or reference Temporary and secret key to integrity information.

The first step to generate the vectors of projection is to create a vector with all its elements set to one and determine the sign of the vector elements pseudorandomly. Next, a mask is applied to Get the projection vector. The mask can be generated responding to human psycho-visual characteristics or other types of requirement.

An example to produce the mask is to create it so that the weight of the coefficients in the projection is the same from the perspective of the JPEG standard; for this the values of the mask are generated as a function of the size ratio of the quantification steps associated with JPEG quantification of each coefficient. It was previously indicated that each coefficient of the 8 \ times8 blocks into which an image is divided into the JPEG standard undergoes a quantification, where the size of the Quantifier step is a function of its position in the block. Therefore, the mask is generated so that when projecting coefficients on the projection vector the mask value be such that the product of the JPEG quantification step of each coefficient for the value of the mask that multiplies it be constant. This match is achieved by assigning in the mask the greater value (for example 1) to the element associated to the coefficient quantified with the smallest step and not making the rest of The elements of the vector are proportional to it. For example, yes a coefficient associated with a step size whose value is twice that of the smallest coefficient would correspond to it in the mask half the value associated with the smallest. Particularizing, if only one coefficient of each block is marked to enter a certain message, the generated mask would be a vector with its elements equal to unity. This way of generate the mask is illustrated in the following example

c_ {quant} = {1,1,2,1}

q = {5,5,6,6).

In the previous case it represents the vector formed by the quantified coefficients, the vector q represents the values of the JPEG quantification matrix with which they are obtained the quantified coefficients. Attending to this, the mask m result

m = {1,1,5 / 6,5 / 6}.

May correspond to projection vector p = {1, -1.5 / 6, -5 / 6}.

Watermark Insertion

In the result of the RES_PROY projections, enter a bit of the messages or a part of the information from integrity. This process is carried out in steps 124 of the Fig. 4, 145 of Fig. 5 and 168 of Fig. 6 for the IP address, Temporary reference and integrity information respectively. The three steps are analogous, differing only in the information that is entered in each one.

In the system proposed in the present invention watermarking techniques with lateral information are used in the encoder, and in the present preferred description a technique based on uniform scalar quantifiers and repeat coding.

The messages to be inserted are represented by binary vectors (their elements can only take values {0,1}) in which each element represents a bit of information:

The message corresponding to the IP address is denoted by and has a length of 32 bits, where this value is denotes in the flowcharts of Fig. 4 and Fig. 7 by NºBITS_IP.

The temporary reference message is denoted by and has a variable length, depending on the coding channel applied in step 133 of Fig. 5. This value of Length is denoted in Fig. 5 and Fig. 8 by No. BITS REF.

Integrity information is also represented by a binary vector that represents the value of a reference message, entering a bit in each macroblock, and therefore its length is equal to the number of macroblocks of the image. This reference message is arbitrary, and must be known by the decoder to verify its presence in the picture.

The objective of the watermarking process that described in this preferred practical embodiment is to code each message or integrity information in a vector and that represents a code word, which will be inserted in the image original.

In the method used in this description, each code word and has L times the length of the binary vector that represents the message to be inserted, with L being the repetition rate used NºCOEF_IP / LONG_IP, NºCOEF_REF / LONGREF or NºCOEF_INT / LONG_INT as appropriate to the address IP, temporary reference or integrity information respectively; in general L will be different for each message or for the integrity information. The process for entering information in the image proposed in this description of the preferred embodiment of the invention comprises three steps: quantification, obtaining error vectors and updating the coefficients. These steps will be described in detail below.

In the first step, to insert a message or the integrity information of length N represented by a binary vector b = (b_ {1}, b_ {2}, ..., b_ {N}) the elements of the code word and, so that for an element with index i within the range (( j -1) \ times + 1, j \ times L ] the corresponding binary vector bit will be used, being 1 < j < N . Where b denotes the vectors or for the case of the messages or for the integrity information. The ith value is calculated as

100

where Δ is the so-called quantification step , which will determine the distortion introduced by the marking process. On the other hand, in the previous expression (1) it represents a pseudo-random value evenly distributed in the range [-1 / 2,1 / 2] and is known only by the encoder and the decoder; In addition, it denotes the result of the i- th RES_PROY projection of the corresponding insertion process in an image as carried out in the insertion module of the IP address (Fig. 4), in that of the IP address (Fig. 5 ) or in that of the integrity information (Fig. 6), and where it represents the quantification operation, which due to the particular structure of the code words is simply given by

101

Obtaining is plotted in the Fig. 12, where circles and squares symbolize points reconstruction representing bits 0 and 1, respectively. Regarding the pseudorandom value, it can obtained from the secret key K or from the temporary reference for the messages and the secret key in the case of integrity information. In general, a value will be used different for each value in order to provide privacy to the marking process. The vector formed by those of each message or Integrity information is represented by for the IP address, for temporary reference, and for information on integrity.

The second step of the marking process consists of in obtaining the quantization error vector d.

Each element is the result of steps 124 (Fig. 4), 145 (Fig. 5) and 168 (Fig. 6).

The third step of the marking process is the update, an operation by which the value of each is dispersed over all the coefficients of the original image that have been used to obtain the value of (which corresponds to a certain RES_PROY value). The update is carried out in steps 125 (Fig. 4), 146 (Fig. 5) and 169 (Fig. 6). It is assumed that the coefficients of the image, which have been used to obtain the value of RES_PROY that resulted in the previous step by means of the projection operation, are arranged in the vector c = (c_ {1}, c_ { 2} ..., c_ {M}) , where M is equal to LONG_IP, LONG_REF or LONG _INT, depending on whether the message inserted corresponds to the IP address, time reference, or integrity information, respectively. Obtaining j -th coefficient marked, denoted by c _ {j} ^ {\ text {*}} is given by the following expression:

102

where d {i} ^ {\ text {*}} = d {i} / M is the element j -th vector corresponding projection (i.e., or), and \ alpha is called compensation factor of distortion , which can take real values in the interval [0,1]. By controlling the value of α, a compromise solution can be reached between the distortion introduced and the robustness of the brand.

To conclude the marking process, the original coefficients are replaced by the coefficients marked c _ {j} ^ {\ text {*}}.

Distortion introduced by JPEG

The digital image resulting from altering the coefficients in the insertion of information is re-encoded according to the JPEG standard. As indicated above the standard JPEG quantifies the coefficients of each 8 \ 8 block that they form the image by the elements of the quantization matrix JPEG, introducing a quantization noise.

It is necessary to know the values of in the process of inserting the image information, so that it can set a minimum step size value \ Delta necessary to insert the information. This generation process of the elements of the code words and described in the expression (1).

The determination of the size of the steps of the quantifiers (y), depends on the minimum value of the size of the quantification step of (q_ {IP, min}, q_ {REF, min}, or q_ {INT, min}); this value divides the coefficients used to Insert each type of information. That way you get the communication established between the encoder 100 and the decoder 200 is possible with a very small distortion. The condition that must be met is

\ Delta_ {IP} \ geq 2q_ {IP, min},

\ Delta_ {REF} \ geq 2q_ {REF, min} Y

\ Delta_ {INT} \ geq 2q_ {INT, min}.

Respecting this condition, the information entered resists JPEG compression at a certain quality factor, where that factor controls. This is clear from the example described below. A factor Δ = 1 is assumed, without projection and the result of the quantization error for the introduction of a bit, whose value is zero, is the maximum possible value d = Δ / 2. The marked coefficient results c ^ {\ text {*}} = c + \ Delta / 2. If the aforementioned condition was not met, the quantified marked value obtained c {quant} ^ {\ text {*}} would be equal to the original quantized unmarked value, therefore in decoder 200 it would produce an error when recovering the value of the bit inserted since
\ hat {\ mathit {c}} = c _ {quant} ^ {\ text {*}} \ times that would be closer to a code word of the subset associated with the value one than the value zero.

Information Extraction

The process of extracting the inserted information is similar to the marking process, and can be broken down into two stages. To explain it, we start with a set of vectors of coefficients c = (c_ {1}, c_ {2}, c_ {M}) , j = 1 ... N , which have been marked to transmit information about of the jth bit of a message that has a total length of N bits, this value corresponding to NºBITS_IP or NºBITS_REF for the case of the IP address and the temporary reference respectively or the number of macroblocks of the image for the integrity information . In addition, the L value is the repetition rate used, that is, COCO_IP / LONG_IP No., COEF_REF / LONG_REF No. and COEF_INT / LONG_INT No. for the IP address, time reference and integrity information respectively.

The first step of the decoding process is to obtain a distance vector VEC-DIS = (v_ {1}, v_ {2}, ..., v_ {N}) . The projection of the vector using the appropriate parameters results in a vector of length L , with 1 <j <N, and we denote the concatenation of said N vectors as s . The value of v j is obtained by the expression

103

The value represents the absolute value of the quantizer error of the subvector with the set of code words representing bit 0 in the jth bit, where the set code words has the form of the vector (\ Delta (t_ {1} + k_ {1}), \ Delta (t_ {2} + k_ {2}), ..., \ Delta (t_ {L} + k_ {L})) , where it is an integer and the pseudorandom values must correspond to those used in the coding phase. Operation (2) is carried out in steps 226 (Fig. 7), 246 (Fig. 8) and 264 (Fig. 9). Depending on the type of message considered, the next step in the decoding process is as described below.

To obtain the jth bit of the IP address (step 217 of Fig. 7), the decision rule is given by

104

that is, a criterion of minimum distance.

The message regarding the temporary reference is Encode against errors using a channel code (e.g. convolutional). In this case, the distance vector is the input to block 235 of Fig. 8, which obtains the vector at the exit binary \ hat {\ mathit {b}} _ {REF}, which represents the message Dear.

In the system proposed in the present invention the detection of temporary alterations in the images processed by him. This happens when you try to modify an image or a sequence of images is replaced. It is implemented using a temporary window composed of references Temporary valid to be checked against the temporary reference extracted from each image. If the temporary reference is within the window, the window is updated with the value of the new Temporary reference If the extracted temporary reference is not valid indicates that that particular image or sequence of images is not valid.

The integrity information is extracted in step 261 of Fig. 9. In this step, the distance between the set of code words associated with a reference message, and the projected version of the integrity coefficients of j - will be calculated. th macroblock. If this distance is less than a certain threshold, then it will be decided that the signal is authentic, and in another case it has been edited.

Device Integration

Note that flowcharts do not use No special syntax, or any programming language. Plus well, they represent the information necessary for a person familiar in this field of technology can manufacture integrated circuits or generate the software running the necessary processes. For example, each function represented by a block or a flowchart can be implemented by a software instruction set, by a digital processor of DSP signal, by an FPGA configurable digital circuit, by a ASIC specific application circuit or any combination of they.

In order to illustrate a possible implementation of watermark insertion methods proposed in the present invention is shown in Fig. 13 a block scheme of a 1300 network digital camera. The camera it comprises a lens 1301 that focuses an image on a sensor of image 1302, an image generating circuit 1303 that uses the image captured by image sensor 1303 causing an image digital encoded according to some standard (eg JPEG) and a control circuit 1304 having as one of its functions the control of obtaining, generating and coding the images, being another of its functions the communication with a network of communications 1306. In addition, control circuit 1304 has the ability to execute operations stored in a memory 1305, where operations are stored necessary to carry out the proposed insertion methods In this invention. Generally, the low calculation capacity of the control circuits present in the current digital cameras network, make the present invention suitable by conjugating perfectly a high degree of security with the need for a low number of operations

Digital watermarking methods for tamper detection proposed in the present invention is could be implemented in a 1400 computer system, as shown in Fig. 14, where such a computer system 1400 it would comprise a processor 1401 and a memory 1403. Such a system of 1400 computing would be connected to devices of a system 1407 surveillance through a 1405 digital communications network. The processor 1401 executes the operations stored in the memory 1403; therefore the described processes that are part of the methods for detecting alterations can be implemented in these computer systems.

As an example, the 1400 computer system it could be a central computer that controls the parameters of the digital cameras that make up the global system or simply a DSP configured to randomly analyze some recordings of a database.

So far, a possible detail has been detailed practical embodiment of the present invention. It is evident for a person prepared in this technical field that there are variants in the State of the art applicable to the practical realization presented. According to this, it should be said that the scope of this invention will be limited only by the claims that accompany the invention and not because of the content of the description preferential.

Claims (16)

1. A digital watermarking system with digital image manipulation detection for video surveillance systems by inserting integrity information and at least one message, characterized in that the watermarking method comprises the steps of: transform the digital image into an image digital transformed, divide the transformed digital image into a plurality of blocks, each block having a plurality of coefficients; obtain a plurality of projected values of message for each of the messages projecting a plurality of predetermined coefficients on a vector dependent on a Secret password; insert each bit of each message at least in one of said projected message values; obtain a plurality of projected values of integrity for said integrity information by projecting a plurality of predetermined coefficients on a vector message dependent, said message dependent vector being derived from at least one of said messages and the key secret; insert each part of the information from integrity in at least one of said projected values of integrity; update each default coefficient used to calculate the projected message values with the values projected message inserts, making the projection of the updated default message coefficients on the vector dependent on the secret key and projected values message inserts are identical; Y update each default coefficient used to calculate the projected integrity values with the values projected inserts of integrity, making the projection of the updated default integrity coefficients over said message dependent vector and the projected values integrity inserts are identical. 2. System according to claim 1, characterized in that the step of inserting each part of the integrity information is calculated by subjecting said projected integrity values to a quantification using a lattice quantizer with a predetermined step size and moving the centroids of the quantifier of latticework with a synchronized scroll vector with at least one of said messages and the secret key; and the step of inserting each bit of each message is calculated by subjecting said projected message values to a quantization using a lattice quantizer with a predetermined step size and moving the centroids of the lattice quantizer with a shift vector synchronized with the secret key . 3. System according to claim 1, characterized in that each plurality of predetermined coefficients used to insert the messages is selected pseudorandomly with the secret key. System according to claim 1, characterized in that each value of said message dependent vector having a specific position in said message dependent vector and each value of the secret key dependent vector having a specific position in the secret key dependent vector they are weighted by a factor, where said factor is derived from said specific position of each value in the vector. 5. System according to claim 4, characterized in that said digital image is encoded with a digital image standard obtaining an encoded digital image, wherein said digital image standard is selected from the group consisting of the JPEG standard and any of the MPEG standards; each coefficient is coded by dividing it by a value from the quantification table of the standard, where said value from the quantification table is a function of the position of the coefficient in the particular block to which it belongs; and the value of each factor that weights a specific value of the dependent vector is a function of the value of the quantification table that divides a specific coefficient in the encoded image, where said concrete coefficient multiplies said concrete value of the dependent vector when projecting in the steps to obtain the pluralities of projected values. 6. System according to claim 1, characterized in that one of said messages is a temporary reference. A system according to claim 1, characterized in that one of said messages is a unique identifier of a device, wherein said device captured said digital image. 8. A digital watermarking system with tamper detection in a digital image with watermark for video monitoring systems by extracting data from said digital image with watermark, where said data is integrity information and at least one message, characterized in that the watermarking method comprises the steps of: transform the digital image into an image digitally transformed, divide the transformed image into a plurality of blocks, each block having a plurality of coefficients; obtain a plurality of projected values for each of said messages projecting a plurality of predetermined coefficients on a vector dependent on a Secret password; extract a plurality of bits from each message of the projected message values; for each message determine a message rebuilt based on the bits extracted from each message; obtain a plurality of projected values of integrity for integrity information by projecting a plurality of predetermined coefficients on a vector message dependent, which is derived from at least one of the reconstructed and secret key messages; extract a plurality of parts of the integrity information of said plurality of values integrity projections; Y determine if the digital image with the mark of water was altered or not analyzing the parts extracted from the integrity information System according to claim 8, characterized in that the step of extracting a plurality of parts of the integrity information is calculated by subjecting said projected integrity values to a quantification using a lattice quantizer with a predetermined step size and moving its centroids of lattice quantizer with a synchronized offset vector with at least one of said reconstructed messages and the secret key; and in the step of extracting a plurality of bits of each message from the projected values, it is calculated by subjecting said projected message value to a quantification using a lattice quantizer with a predetermined step size by moving its lattice quantizer centroids with a vector of synchronized scrolling with the secret key. 10. System according to claim 8, characterized in that each plurality of predetermined coefficients used to decode the reconstructed message is selected pseudorandomly with the secret key. A system according to claim 8, characterized in that each value of said message dependent vector having a specific position in said message dependent vector and each value of the dependent vector of the secret key having a specific position in the dependent vector of the secret key they are weighted by a factor, where said factor derives from said specific position of each value in the vector. 12. System according to claim 11, characterized in that said digital image is encoded with a digital image standard obtaining an encoded digital image, wherein said digital image standard is selected from the group consisting of the JPEG standard and any of the MPEG standards; each coefficient is coded by dividing it by a value from the quantification table of the standard, where said value from the quantification table is a function of the position of the coefficient in the particular block to which it belongs; and the value of each factor that weights a specific value of the dependent vector is a function of the value of the quantification table that divides a specific coefficient in the encoded image, where said concrete coefficient multiplies said concrete value of the dependent vector when projecting in the steps to obtain the pluralities of projected values. 13. System according to claim 8, characterized in that one of said messages is a temporary reference that is used to decide whether the digital image with watermark was obtained in a valid period of time or underwent temporary manipulation. 14. System according to claim 8, characterized in that one of said messages is an identifier of a device, wherein said device captured said image; said unique identifier is used to verify whether the digital image was altered or not. 15. A digital network camera, which is characterized in that it comprises: a lens to focus an image on a image sensor; an image generation circuit to generate a digital image from an image captured by said image sensor; a control circuit to control the communications between the digital camera and the network to which it is connected; and said control circuit is prepared to to carry out a method according to any of the claims 1 to 7. 16. A computer system including a processor and a memory, wherein said computer system is integrated into a video monitoring system, where the computer system is characterized in that it is prepared to carry out a method according to any of the claims from 8 to 14.