Method of detecting and extracting a watermark

Abstract

A method of detecting and extracting a watermark (digital watermarking) in a digital signal. The watermark is composed by the repetition of elementary components of a load to be transmitted. At least one repetition of an elementary component in the digital signal is detected. At least one quality attribute is assigned to each repetition of at least one elementary component. The repetitions of an elementary component are sorted according to a function of the quality attributes of these repetitions. Certain repetitions, dubbed individualities, of an elementary component are selected. The selected repetitions are fused.

Claims

1. A method of detecting and extracting a watermark, a digital watermarking, in a digital signal, said watermark being composed by a repetition of elementary components, said method comprising the steps of: detecting at least one repetition of an elementary component in the digital signal to provide a data sample dubbed an individuality of said elementary component, said at least one repetition being a binary bit; determining at least one quality attribute of each repetition of at least one elementary component, said at least one quality attribute being a bit error rate, a reliable repetition having a lower bit rate than an erroneous repetition; classifying the repetitions of said elementary component according to said at least one quality attribute of these repetitions to identify erroneous repetitions; selecting the repetitions, dubbed individualities, of said elementary component according to a classification of the repetitions to discard the erroneous repetitions and to keep reliable repetitions of said elementary component; and merging the reliable repetitions by a weighted summation, a weight of each reliable repetition being determined as a function of said at least one quality attribute of said each reliable repetition, to detect and extract the watermark in the digital signal.

2. The method as claimed in claim 1, wherein the quality attribute of each repetition of said elementary component is the reliability of this repetition.

3. The method as claimed in claim 1, further comprising the step of classifying the repetitions of one and the same elementary component by decreasing bit error rate.

4. The method as claimed in claim 1, further comprising the step of discarding the erroneous repetitions from a final process of determining a value to be retained for the elementary component.

5. The method as claimed in claim 1, wherein the elementary components are binary.

6. The method as claimed in claim 1, wherein the elementary components are real quantities, values of the elementary components are in a set of real numbers.

7. The method as claimed in claim 1, further comprising a step of grouping the classification of the repetitions into two classes: a class of winning individualities and a class of minor individualities.

8. The method as claimed in claim 7, further comprising the step of evaluating a boundary between domains of the winning individualities and of the minor individualities by a statistical computation on a total or partial population of the repetitions of the elementary components.

9. The method as claimed in claim 7 wherein a merger of the winning individualities arising from the repetition of one and the same elementary component is carried out by summation.

10. The method as claimed in claim 1, wherein the digital signal is representative of a still image.

11. The method as claimed in claim 1, wherein the digital signal is representative of a sequence of moving images.

12. The method as claimed in claim 1, wherein the digital signal is representative of a sound sequence.

13. A digital watermark decoding device for detecting and extracting a watermark, a digital watermarking, in a digital signal, said watermark being composed by a repetition of elementary components, the digital watermark decoding device detects at least one repetition of an elementary component in the digital signal to provide a data sample dubbed an individuality of said elementary component; determines at least one quality attribute of each repetition of at least one elementary component, said at least one quality attribute being a bit error rate, a reliable repetition having a lower bit rate than an erroneous repetition; classifies the repetitions of said elementary component according to said at least one quality attribute of these repetitions to identify erroneous repetitions; selects the repetitions, dubbed individualities, of said elementary component according to a classification of the repetitions to discard the erroneous repetitions and to keep reliable repetitions of said elementary component; and merges the reliable repetitions by a weighted summation, a weight of each reliable repetition being determined as a function of said at least one quality attribute of said each reliable repetition, to detect and extract the watermark in the digital signal.

Description

DETAILED DESCRIPTION OF AN EMBODIMENT OF THE INVENTION

(1) For the subsequent description, the following terms are defined:

(2) A set of repetitions of the elementary components forming a message (or load) to be inserted within a cover signal, here a digital image, is dubbed a watermark. The data sample resulting from the detection and from the reading in a zone of the image of a repetition of an elementary component of the watermark is dubbed individuality.

(3) As has been seen, the realization of a system for inserting a watermark into a cover signal (digital watermarking of images) requires the fine tuning of two categories of methods:

(4) 1) those which make it possible to insert a message into an image or another file (for example digital audio);

(5) 2) those which make it possible to exploit the marks, stated otherwise to detect and determine the elements of the message buried in these images.

(6) It is clear that the method of inserting a watermark into an image or a sequence of images (or another file, for example digital audio) is, a priori, intended to be used in a systematic manner on all the images generated by a user. On the other hand, the method detecting and reading a message buried in an image is intended to be used much more rarely, typically when a doubt exists as regards the real provenance of an image.

(7) Here, images is intended to mean either single images or sequences of images, for example video sequences. More generally, the method remains usable without significant modification for the processing of coded messages within digital audio files.

(8) The invention proposes a particular method of organizing the detection and extraction of the watermark values constituting the buried message.

(9) In the present exemplary embodiment, the invention is implemented in the form of a software application executed on a computer, for example of PC type.

(10) This software application receives as input one or more digital or digitized images liable to contain a watermark to be detected and read. The software application provides as output the content of the message detected if such a message has been detected.

(11) It is assumed here that the images have been coded (that is to say that a message has been inserted thereinto) with a method using a redundancy of the message to be inserted into the image. It is assumed that the message has repeated a certain number of times in the image and/or the sequence of images to ensure the transmission thereof through a disturbing environment, especially a highly noisy one

(12) The case is considered where the buried message is made up of a sequence of elements forming part of an alphabet A. These may especially but not limitingly entail binary values or real or complex numbers, or a combination of these values.

(13) In a step 101, the signal to be analyzed (watermarked signal) undergoes an elementary detection which produces the individualities, that is to say samples of data arising from the repetition of an elementary component of the watermark.

(14) In a step 102, the individualities are temporarily stored to facilitate their use by the following processing steps.

(15) In a step 103, characteristics of quality of each individuality, said characteristics also being called quality attributes, are evaluated.

(16) In a step 104, statistics are performed on the quality attributes so as to determine dynamically if necessary the operating parameters of step 105.

(17) In a step 105 the individualities are sorted into at least two classes.

(18) In steps 106 and 107, the individualities are temporarily stored to facilitate their use by the following processing.

(19) In a step 108, the individualities associated with one and the same elementary component of the message are used to take a decision about the value of the elementary component (value on the previously cited alphabet A).

(20) In an implementation variant, the cover signal is an image (or a sequence of images) and the image, once coded, comprises a plurality of watermarked blocks disposed at diverse places in the image. In this case, the method comprises, for example, steps such as listed hereinbelow.

(21) In a step, the location of a watermarked block is determined in the image.

(22) In a subsequent step, samples of data (individuality of a repetition within the sense defined above) are extracted from the image. These steps of determining and extracting an individuality are carried out by techniques known per se to the person skilled in the art.

(23) In a subsequent step, a quality attribute is assigned to each individuality extracted. A quality attribute (scalar value) is evaluated for each individuality (repetition of a message element). The individualities (repetitions) are classified by decreasing attribute value. The individualities having attributes of overly low quality are discarded from the final process of determining the value to be retained for the elementary component of the message.

(24) In a subsequent step, there is performed a sorting of the individualities according to this quality attribute and the statistics performed on the population of the values taken by the present attributes.

(25) The selection of the individualities is performed by way of a classifying according to the individual reliabilities and then by the grouping into at least two classes; the class A of the winning samples and a class B of the minor samples.

(26) In this case, more particularly, the boundary between the domains is evaluated by a statistical computation on the populations consisting of the values of individual quality.

(27) The merger of the individualities arising from the repetition of one and the same elementary component is carried out by summation.

(28) The merger of the individualities arising from the repetition of one and the same elementary component is carried out by weighted summation, the weighting being determined by a function of the quality of each sample.

ADVANTAGES

(29) The watermark detection method, such as set forth, applies to the reconstruction of a message that has been repeated a certain number of times. Its use makes it possible to transform the repetitions, highly erroneous, into a single more reliable message, with a more considerable error tolerance capacity than the conventional majority voting or decoding schemes.