WO2019113810A1

WO2019113810A1 - Method and apparatus for embedding and extraction detection of three-dimensional blind watermark in local spherical coordinate system

Info

Publication number: WO2019113810A1
Application number: PCT/CN2017/115765
Authority: WO
Inventors: 彭翔; 王启垒; 何文奇; 刘晓利
Original assignee: 深圳大学
Priority date: 2017-12-13
Filing date: 2017-12-13
Publication date: 2019-06-20

Abstract

A method and apparatus for embedding and extraction detection of a three-dimensional blind watermark in a local spherical coordinate system, applicable to the field of three-dimensional digital watermarks. An embedding part (S1) comprises: first preprocessing an original watermark image to be embedded to obtain an energy value and an index value to be embedded, then filtering candidate vertices meeting a condition from a three-dimensional model of a watermark to be embedded, establishing local spherical coordinate systems of the candidate vertices, and finally embedding index information and the energy value by performing attribute substitution under the constraint of invisibility, thereby completing an embedding process of the watermark image; and an extraction detection part (S2) comprises: first filtering possible embedded vertices, establishing local spherical coordinate systems of the possible embedded vertices to obtain spherical coordinate values, extracting the index value and the energy value, then performing reverse transformation to return back to a spatial domain to obtain an extracted watermark image, and finally performing relevant calculation on the extracted watermark image and the original watermark image to determine whether the watermark is present. The invisibility embedding and blind extraction for a three-dimensional model visual blind watermark is achieved.

Description

Three-dimensional blind watermark embedding and extraction detection method and device in local spherical coordinate system

Technical field

The invention belongs to the field of three-dimensional digital watermarking, and in particular relates to a three-dimensional blind watermark embedding and extracting detection method and device in a partial spherical coordinate system.

Background technique

In recent years, with the development of 3D imaging technology, 3D digital models are increasingly appearing in people's field of vision as a new form of digital media. Especially with the development of computer graphics, virtual reality, 3D games, 3D display, etc., 3D digital models are becoming an important form of digital media. The illegal possession, theft, modification, and dissemination of the three-dimensional digital model are also increasing, and the intellectual property protection of the three-dimensional digital model is imminent.

As an effective technical means of copyright protection, digital watermark can also be used to protect intellectual property rights of 3D digital models. The research on more mature digital watermarking technology is mostly concentrated on one-dimensional or two-dimensional digital carriers such as audio, image and video. There are few researches on watermarking of three-dimensional models. According to whether the original model is used to detect the watermark, the 3D model watermark can be divided into a blind watermark and a non-blind watermark. The latter needs the original model when detecting the watermark, while the former does not need it. Under the current 3D model retrieval technology, the non-blind The watermark value is not large, and the blind watermark technology has practical application value. In addition, the watermark is further divided into a checkable watermark, a readable watermark, and a visible watermark according to the content. The detection result of the watermark detectable is a binary form of 0 or 1. The detection result of the readable watermark is a string form, and the detection result of the visible watermark is an image form. Undoubtedly, the detection result of the visible watermark is more intuitive and meaningful.

Compared with image watermarking, audio watermarking and video watermarking, the carrier of the three-dimensional model watermark is different. The vertices of the 3D model are irregular sampling, and there is no natural ordering, which makes no direct mathematical tools to process them, so the watermark embedding is very difficult. Compared with 3D blind watermark, 3D non-blind watermark can also extract watermark by means of resampling and realignment, while blind watermark can only rely on its own watermark. Watermark extraction, embedding difficulty and extraction difficulty is more, which is one of the reasons why most 3D model watermarks use non-blind detection.

Under the current three-dimensional model retrieval technology, the research on visualized blind watermarking of 3D model is very difficult, but it has important practical significance.

Summary of the invention

The invention provides a method and a device for embedding and extracting a three-dimensional blind watermark in a partial spherical coordinate system, aiming at realizing invisibility embedding of a watermark of a three-dimensional digital model, realizing online blind extraction of the watermark after embedding, and the watermark content is meaningful Visual watermark.

The present invention provides a three-dimensional blind watermark embedding and extraction detection method in a partial spherical coordinate system, the method comprising: a blind watermark embedding step S1 and a blind watermark extraction detecting step S2;

The blind watermark embedding step S1 includes:

Step S11, pre-processing the original watermark image to be embedded, and obtaining the attribute value (L, J, I) of the watermark element of the original watermark image;

Where L is the row index code value, J is the column index code value, and I is the element value, ie the energy value;

Step S12: Filter all candidate vertices satisfying the screening condition from the three-dimensional model to be embedded in the watermark according to the first preset screening condition;

Step S13, establishing a local spherical coordinate system under the custom for the candidate vertices, and determining a spherical coordinate value of each of the candidate vertices

among them,

θ, r are three attribute values of each candidate vertex satisfying the screening condition;

Step S14, comparing the attribute value of the watermark element with the attribute value of all candidate vertices satisfying the screening condition, searching for a candidate vertex that has the smallest difference from the attribute value of the watermark element as a point embedded in the watermark element, and changing the candidate The position of the vertex in its spherical coordinate system, making it

The value of θ is equal to the row index code value L and the column index code value J of the watermark element, respectively, and the r value thereof is equal to the element value I of the watermark element to embed the index code value and the energy value of the watermark element; The candidate vertices are culled in all candidate vertices satisfying the screening condition, and all the watermark elements of the original watermark image are traversed until the above is completely embedded;

The blind watermark extraction detecting step S2 includes:

Step S21: Filter all possible embedded vertices satisfying the screening condition from the three-dimensional model to be detected according to the first preset screening condition;

Step S22, establishing a local spherical coordinate system under the custom embedding vertices, and determining a spherical coordinate value of each of the possible embedded vertices

among them,

θ, r are three attribute values of each possible embedding vertex that satisfy the screening condition;

Step S23, combining the attribute values of the possible embedding vertices with the index value and the energy value extraction of each of the possible embedding vertices satisfying the screening condition in the three-dimensional model, to obtain a matrix S4;

Step S24, inversely transforming the matrix S4 back into the airspace to obtain an extracted watermark image;

Step S25: Calculate a correlation between the extracted watermark image and the original watermark image by using a watermark correlation calculation formula, and if the correlation is greater than an experience threshold, prove that the three-dimensional model includes a watermark.

The invention also provides a three-dimensional blind watermark embedding and extracting detecting device in a partial spherical coordinate system, the device comprising: a blind watermark embedding module and a blind watermark extraction detecting module;

The blind watermark embedding module includes:

a pre-processing sub-module, configured to pre-process the original watermark image to be embedded, to obtain an attribute value (L, J, I) of the watermark element of the original watermark image;

a first screening sub-module, configured to filter, from the three-dimensional model of the watermark to be embedded, all candidate vertices satisfying the screening condition according to the first preset screening condition;

a first coordinate system establishing submodule, configured to establish a local spherical coordinate system under the custom vertices, and determine a spherical coordinate value of each of the candidate vertices

among them,

a watermark embedding sub-module, configured to compare a property value of the watermark element with an attribute value of all candidate vertices satisfying a filter condition, and search for a candidate vertex that has the smallest difference from the attribute value of the watermark element as a point at which the watermark element is embedded , changing the position of this candidate vertex in its spherical coordinate system, making it

The blind watermark extraction detection module includes:

a second screening sub-module, configured to filter, from the three-dimensional model to be detected, all possible embedded vertices satisfying the screening condition according to the first preset screening condition;

a second coordinate system establishing submodule, configured to establish a local spherical coordinate system under the custom embedding vertices, and determine a spherical coordinate value of each of the possible embedded vertices

among them,

An extraction submodule, configured to perform an index value and an energy value extraction on the possible embedded vertex satisfying the filter condition in the three-dimensional model in combination with the attribute value of the possible embedding vertex, to obtain a matrix S4;

An inverse transform sub-module, configured to inverse transform the matrix S4 back into the airspace to obtain an extracted watermark image;

The determining sub-module is configured to calculate a correlation between the extracted watermark image and the original watermark image by using a watermark correlation calculation formula, and if the correlation is greater than an experience threshold, prove that the three-dimensional model includes a watermark.

Compared with the prior art, the present invention has the beneficial effects that the method and device for three-dimensional blind watermark embedding and extracting detection in a partial spherical coordinate system provided by the present invention include a watermark embedding portion and a watermark extraction detecting portion, wherein the watermark embedding portion The processing process is: pre-processing the embedded original watermark image, obtaining the energy value and the index value to be embedded, and then screening out the candidate vertices satisfying the condition from the three-dimensional model of the watermark to be embedded according to the first preset screening condition, and The local spherical coordinate system of the candidate vertices is established. Finally, the attribute replacement under the constraint of invisibility is used to embed the index information and the energy value, that is, the embedding process of the watermark image is completed; the processing process of the watermark extraction detection part is: first, according to the first The preset filter condition selects the three-dimensional model to be detected and may embed the vertices to establish a partial spherical sitting that may embed the vertices. The target system obtains the spherical coordinate value, and extracts the index value and the energy value, then inversely transforms back into the spatial domain to obtain the extracted watermark image, and finally performs correlation calculation on the extracted watermark image and the original watermark image to determine whether the three-dimensional model contains a watermark; Compared with the prior art, the invention realizes the invisibility embedding of the watermark of the three-dimensional digital model, and does not need to be repeatedly embedded, can realize the online blind extraction of the watermark after embedding, and can resist the affine transformation attack and the shear attack, and the watermark The content is a meaningful visual watermark, which has great practical value.

DRAWINGS

1 is a schematic flow chart of a method for embedding and extracting a three-dimensional blind watermark in a partial spherical coordinate system according to an embodiment of the present invention;

2 is a schematic flowchart of a blind watermark embedding step according to an embodiment of the present invention;

3 is a binary watermark image of the letter L provided by the embodiment of the present invention;

4 is a watermark image obtained by preprocessing an original watermark image provided by an example of the present invention;

FIG. 5 is a schematic diagram of a three-dimensional digital model of a watermark to be embedded according to an embodiment of the present invention; FIG.

6 is a three-dimensional digital model embedded in a watermark according to an embodiment of the present invention;

7 is a schematic flowchart of a blind watermark extraction detection step according to an embodiment of the present invention;

FIG. 8 is a watermark image obtained by performing watermark extraction on a three-dimensional model embedded in a watermark according to an embodiment of the present invention;

FIG. 9 is a schematic block diagram of a three-dimensional blind watermark embedding and extracting detection apparatus in a partial spherical coordinate system according to an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention provides a new blind watermarking method for transform domain coding in a partial spherical coordinate system, including embedding Into the part and the extraction detection part, the main implementation idea is: embedding the index value and energy value of the spectrum energy matrix of the original watermark image into the custom coordinates of the local spherical coordinate system of the model vertex, and then extracting the local spherical coordinate system The index value and the energy value are inversely transformed back into the spatial domain to obtain the extracted watermark image, and the watermark image is compared with the original watermark image to determine whether the watermark is included; wherein the transform domain of the embedded portion is a two-dimensional discrete cosine transform The transform domain of the extracted detection portion is a two-dimensional inverse discrete cosine transform, and the watermark image is a binary image.

The following is a detailed description of the three-dimensional blind watermark embedding and extraction detection method in the local spherical coordinate system, as shown in Figure 1, including: blind watermark embedding step S1 and blind watermark extraction detection step S2;

Wherein, as shown in FIG. 2, the blind watermark embedding step S1 includes:

Where L is the row index coded value, J is the column index coded value, and I is the element value, ie, the energy value.

Specifically, in the embodiment of the present invention, a picture with a size of W1×W2 and a content of the letter L is selected as the original watermark picture, which is denoted as S; as shown in FIG. 3, the original watermark picture is obtained, and FIG. 4 is obtained by preprocessing the original watermark picture. image.

The step S11 specifically includes:

Step S111, using a two-dimensional discrete cosine transform formula to perform two-dimensional discrete cosine transform on the original watermark image to be embedded, to obtain a spectral matrix S1 of the discrete cosine transform domain of the original watermark image;

The two-dimensional discrete cosine transform formula is:

Where u=0,1,...,W1-1;v=0,1...,W2-1;

Wherein W1 represents the length of the original watermark picture, W2 represents the width of the original watermark picture, the original watermark picture is a binary watermark picture, and the matrix of the original watermark picture is represented as f(x, y), S1(u, v A spectral matrix obtained by discrete cosine transform of a matrix f(x, y) of the original watermark image, the spectral matrix being a real matrix with positive and negative coefficients.

Step S112, extracting the positive and negative coefficient matrix of the spectrum matrix S1 as the recovery key K1, performing absolute value on the spectrum matrix S1, and performing normalization processing to obtain a matrix S2, and the maximum value of the matrix S2. As the recovery key K2;

Specifically, in the embodiment of the present invention, the spectrum coefficient matrix is used as the secret key K1, the normalization coefficient is used as the secret key K2, and the scrambling parameter is used as the secret key K3.

Step S113, performing Arnold scrambling on the matrix S2 by using a scrambling matrix to obtain a matrix S3, where the scrambling matrix is:

Where a and b are set parameters, k is the number of iterations, W2 is the width of the original watermark image, a, b, k, W2 are used as the recovery key K3; each element in the matrix S3 has three Attribute values (m, n, I), where m is the row index value, n is the column index value, and I is the element value, ie the energy value;

Step S114, using the following formula to respectively encode the attribute values m, n of each element in the matrix S3 into an angle value, to obtain a new attribute value (L, J, I) of each element of the matrix S3;

Where α and β are preset constants, β is a value between 0 and 1, and considering the invisibility of the watermark, the alpha value should be set to a value between 0 and 0.5, and this value is embedded. After the watermarking, the visual effect of the 3D model is minimally determined as the basis for determination; thus, the new attribute value (L, J, I) of each element of the matrix S3 is obtained, where L is the row index encoding value and J is the column index encoding value. I is an energy value, L _i represents an index code value of the i-th row, and J _j represents an index code value of the j-th column.

Specifically, as shown in FIG. 5, a three-dimensional model to be embedded in the watermark; the candidate vertex is a vertex of the three-dimensional model to be embedded in the watermark, and the watermark information is embedded in the attributes of the vertex.

Specifically, the first preset screening condition includes: a first screening condition, a second screening condition, and a third screening condition, wherein the first screening condition is a centroid neighbor point of a vertex and a centroid of a ring neighbor point The distance maximum ratio needs to be smaller than the threshold A, and the second screening condition is that the normal angle under two different definitions of the centroid of a ring neighbor point needs to be greater than the threshold B, and the third screening condition is that the candidate vertex is not Neighbor point. The reason for setting the second screening condition is that when the local spherical coordinate system is established for the vertices of the three-dimensional model, it is required that the normal n _a and the normal n _{t of the} vertices of each three-dimensional model cannot overlap or cannot coincide. For each vertex in the 3D digital model, if the attribute information is not considered, the triangular mesh model M can be expressed as M = {V _M , K _M }, where V _M = {0, 1, 2, ... M-1} Is the set of vertices of M, M represents the number of vertices, K _M is the set of all topological connection relations of _M , and the elements of K _M are divided into three types: vertex v={i}, edge e={i,j}, Face f = {i, j, k}. If the edge {i,j}∈K _M , the vertices {i} and {j} are mutually called neighbors; the one-ring neighbor of the vertex {i} is defined as N(i)={j|{i,j}∈K _M }.

The step S12 specifically includes:

Step S121, the centroid link neighbor point three-dimensional model is calculated to be embedded watermarks vertices V _i to Z _i, are connected to a ring neighbor point Z _i and V _i, and to calculate the appropriate distance, if each of the distance If the maximum ratio is greater than the threshold A, the vertex is culled, and all the vertices of the three-dimensional model are traversed by the operation to obtain all the first candidate vertices satisfying the condition;

Step S122, calculating normal directions n _a and n _t for the first candidate vertex, if the angle between the normal n _a and the normal n _t is less than the threshold B, then the first candidate vertex is eliminated, for all The first candidate vertex traverses this operation to obtain all second candidate vertices satisfying the condition;

Wherein the normal n _a is defined as:

Connect a circle of neighbor points of the centroid point Z _i and V _i to obtain a number of new triangles with the centroid point as the apex, denoted as T(Z _i ), and the number of triangles is denoted as N, the new triangle The normal is denoted as n _j , and the mean of all the normals of the triangle is taken as the normal n _a of the centroid point Z _i , namely:

The normal n _t is defined as:

Calculating another normal n _t of the centroid according to the square of the normal n _j and the area S _j of each of the triangles, namely:

Step S123, it is checked whether all the second candidate vertices satisfying the condition have vertices which are mutually neighbors, and if so, cull them, and the remaining second candidate vertices are candidate vertices of the watermark embedding.

among them,

Specifically, step S13 needs to establish respective local spherical coordinate systems for the watermarkable candidate vertices obtained in step S12, and determine the spherical coordinate values of each candidate vertex. The general spherical coordinate system and the rectangular coordinate system are transformed as follows:

r=sqrt(x ² +y ² +z ² )

θ=arccos(z/r)

Specifically, in step S13, the spherical coordinate value

Finding the spherical coordinate value of the candidate vertex for the customized encoded spherical coordinate value

The formula is as follows:

r=sqrt((xx ₀ ) ² +(yy ₀ ) ² +(zz ₀ ) ² )/ρ

θ=arccos(|zz ₀ |/r)

Where (x, y, z) is the coordinate value of the candidate vertex in the Cartesian coordinate system, (x ₀ , y ₀ , z ₀ ) is the centroid point coordinate of the ring neighbor of the candidate vertex, and ρ is the centroid point to the neighbor point The mean of the distance.

Specifically, for the candidate vertex V _i , its spherical coordinate system is determined by the one-ring neighbor centroids Z _i , n _a , n _t obtained in step 12, and the centroid Z _{i is taken} as the origin of the spherical coordinate system, and the centroid normal is n _a As the Z axis of the spherical coordinate system, the normal n _t is projected into the plane defined by the centroid Z _i and the normal n _a , and the obtained projection vector is taken as the X axis of the spherical coordinate system. The angle between the vector of the centroid Z _i to the vertex V _i and the Z axis of the spherical coordinate system is recorded as

The forward angle referred to the X-axis spherical coordinates is θ _i, the centroid Z _i to vertex V _i is the distance from the centroid Z _i to the ratio of the distance of a ring neighbor point vertex V _i mean is denoted r _i. Traversing all candidate vertices to perform the above operations, thereby obtaining the spherical coordinate values under the custom of all candidate vertices

The spherical coordinate value of each candidate vertex can be regarded as its three attributes, these three attributes do not change with the change of the coordinate system, and have strong stability.

Specifically, the step S14 is an embedded part of the watermark. The embedding of the watermark image needs to first consider the visual difference between the 3D model and the original 3D model after embedding the watermark, so additional constraints are needed when embedding the watermark. From step 11 and step 13, respectively, the attribute values of the original watermark image and the three attribute values of the candidate vertex are obtained, and the embedding is performed according to the size of the watermark element attribute value I, in the order of whether it is large to small or small to large. You can set it yourself.

Specifically, the essence of the step of embedding the watermark is to determine which candidate vertex is embedded by the watermark element by the attribute value difference, and then determine the coordinate value of the candidate vertex after the position is changed by the watermark element attribute value; wherein, by changing the candidate vertex in it Position in the spherical coordinate system, making it

The value of θ (under the custom) is equal to the index code value L, J of the watermark element, respectively, such that its r value (under the custom) is equal to the watermark element value I, that is, the distance between the vertex and the centroid point of mass and the centroid to one The ratio of the mean value of the ring neighbor point distance is equal to the watermark element value I, thereby realizing the purpose of embedding the index information and the energy value. After the embedding is completed, the candidate vertex is removed from the candidate vertex, and the watermark element is traversed to perform the above operation until fully embedded; The latter three-dimensional model is shown in Figure 6.

Wherein, as shown in FIG. 7, the blind watermark extraction detecting step S2 includes:

Specifically, the step S21 includes:

Step S211, the centroid point of a neighboring ring vertex V _i of the three-dimensional model calculated Z _i to be detected, are connected to a ring neighbor point Z _i and V _i, and to calculate the appropriate distance, if the respective distance The maximum ratio is greater than the threshold A (the threshold set when the watermark is embedded), then the vertex is culled, and all the vertices of the three-dimensional model are traversed by the operation to obtain all the first possible embedded vertices satisfying the condition;

Step S212, calculating normals n _a and n _t under two different definitions for the first possible embedded vertex, if the angle between the centroid normal n _a and the normal n _t of the vertex-ring neighbor point is less than the threshold B (the threshold set when the watermark is embedded), then culling the first possible embedded vertex, traversing the operation for all the first possible embedding vertices, and obtaining all the second possible embedding vertices satisfying the condition;

Wherein the normal n _a is defined as:

Connect the centroid neighbor points of the centroid point Z _i and Vi to obtain a number of new triangles with the centroid point as the apex, denoted as T(Z _i ), and the number of triangles is denoted by N, the method of the new triangle The direction is denoted as n _j , and the mean of all the normals of the triangle is taken as the normal n _a of the centroid point Z _i , namely:

The normal n _t is defined as:

Calculating another normal n _t of the centroid according to the normal n _j and the area S _{j of} each of the triangles, namely:

Step S213, it is checked whether all the second possible embedding vertices satisfying the condition have vertices which are mutually neighbors, and if so, cull them, and the remaining second possible embedding vertices are the finally selected possible embedding vertices.

among them,

Specifically, step S22 is to establish a local spherical coordinate system and acquire a spherical coordinate value. From step 21, it is possible to embed the vertices. Step 22 needs to establish respective spherical coordinate systems for these possible embedding vertices, and determine the spherical coordinate values of each possible embedding vertex, and the spherical coordinate values are the values of the custom encoding when the watermark is embedded. . For the possible embedding of the vertex V _i , the establishment of the local spherical coordinate system and the acquisition of the spherical coordinate value are as follows in step S13, thereby obtaining the possible embedding of the vertex.

Specifically, the step S23 includes:

Step S231, using the following formula to attribute the attribute value of the possible embedding vertex

Perform anti-coding;

Step S232, establishing a zero matrix S4 of the same size as the matrix S3, each matrix element has three attributes (m, n, 0), m and n are row and column values, respectively, and the possible vertices are embedded.

When the value of θ satisfies the following inequality, the r value of the possible embedding vertex is replaced by the zero value of the m row and n column of the matrix S3; all possible embedding vertices are traversed to find the vertex satisfying the index threshold condition, and the r value is replaced Corresponding to the zero value of the matrix S4 of the index, a new matrix S4 is obtained;

Where λ1 and λ2 respectively represent a set threshold value, which satisfies the above inequality

The value of θ is the index value, and the corresponding r value is the energy value.

Specifically, the step S24 includes:

Step S241, using the recovery key K3 in step S113 as a restore key, Arnold reduction of the matrix S4, to obtain a reduction matrix S5;

Step S242, multiplying the elements in the reduction matrix S5 by the recovery key K2 to obtain a matrix S6;

Step S243, using the recovery key K1 in step S112 to multiply the matrix S6, thereby obtaining a spectrum matrix S7;

Step S244, performing a two-dimensional inverse discrete cosine transform on the spectrum matrix S7 by using a two-dimensional inverse discrete cosine transform formula to obtain a matrix S8, where the matrix S8 is an extracted watermark image, and the two-dimensional inverse discrete cosine transform formula as follows:

Where u=0,1,...,W1-1;v=0,1...,W2-1;

Wherein, S8(x, y) represents a watermark image matrix S8 obtained by inverse discrete cosine transform from the extracted spectral matrix S7(u, v); S8 is as shown in FIG.

Specifically, in the step S25, the watermark correlation calculation formula is:

Where W1 represents the length of the original watermark picture, W2 represents the width of the original watermark picture, S(i,j) represents the matrix of the original watermark picture, S(i,j) represents the matrix of the extracted watermark picture, and σ represents the watermark correlation Degree; when the value of σ is greater than the empirical threshold, the model is considered to embed a watermark, and when it is less than the empirical threshold, no watermark embedding is considered.

The invention also provides a three-dimensional blind watermark embedding and extraction detecting device in a partial spherical coordinate system. As shown in FIG. 9, the device comprises: a blind watermark embedding module 1 and a blind watermark extraction detecting module 2.

The blind watermark embedding module 1 includes:

The pre-processing sub-module 11 is configured to pre-process the original watermark image to be embedded, and obtain the attribute value (L, J, I) of the watermark element of the original watermark image;

a first screening sub-module 12, configured to filter, from the three-dimensional model of the watermark to be embedded, all candidate vertices satisfying the screening condition according to the first preset screening condition;

a first coordinate system establishing sub-module 13 configured to establish a local spherical coordinate system under the customization of the candidate vertices, and determine a spherical coordinate value of each of the candidate vertices

among them,

The watermark embedding sub-module 14 is configured to compare the attribute value of the watermark element with the attribute value of all the candidate vertices satisfying the filtering condition, and search for a candidate vertex that has the smallest difference from the attribute value of the watermark element as the watermark element embedded in the watermark element. Point, change the position of this candidate vertex in its spherical coordinate system, making it

The blind watermark extraction detection module 2 includes:

a second screening sub-module 21, configured to filter, from the three-dimensional model to be detected, all possible embedded vertices satisfying the screening condition according to the first preset screening condition;

a second coordinate system establishing sub-module 22, configured to establish a local spherical coordinate system under the custom embedding vertices, and determine a spherical coordinate value of each of the possible embedded vertices

among them,

The extraction sub-module 23 is configured to perform an index value and an energy value extraction on the possible embedded vertex satisfying the filter condition in the three-dimensional model in combination with the attribute value of the possible embedding vertex, to obtain a matrix S4;

An inverse transform sub-module 24, configured to inverse transform the matrix S4 back into the airspace to obtain an extracted watermark image;

The determining sub-module 25 is configured to calculate a correlation between the extracted watermark image and the original watermark image by using a watermark correlation calculation formula, and if the correlation is greater than an experience threshold, prove that the three-dimensional model includes a watermark.

The invention provides a method and device for three-dimensional blind watermark embedding and extracting detection in a partial spherical coordinate system, and combines the idea of airspace and frequency domain to realize invisibility embedding of watermark (so-called invisibility embedding refers to model embedding After the watermark, it does not cause obvious visual difference of the model) and the blind detection of the watermarked 3D model, without the participation of the original 3D model, and can resist the affine attack and the cut attack, and the watermark is a visual watermark, which has Great practical value.

The above is only the preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection of the present invention. Within the scope.

Claims

A method for detecting and embedding a three-dimensional blind watermark in a local spherical coordinate system, the method comprising: a blind watermark embedding step S1 and a blind watermark extraction detecting step S2;

The blind watermark embedding step S1 includes:

Step S11, pre-processing the original watermark image to be embedded, and obtaining the attribute value (L, J, I) of the watermark element of the original watermark image;

Where L is the row index code value, J is the column index code value, and I is the element value, ie the energy value;

Step S12: Filter all candidate vertices satisfying the screening condition from the three-dimensional model to be embedded in the watermark according to the first preset screening condition;

Step S13, establishing a local spherical coordinate system under the custom for the candidate vertices, and determining a spherical coordinate value of each of the candidate vertices
among them,
θ, r are three attribute values of each candidate vertex satisfying the screening condition;

Step S14, comparing the attribute value of the watermark element with the attribute value of all candidate vertices satisfying the screening condition, searching for a candidate vertex that has the smallest difference from the attribute value of the watermark element as a point embedded in the watermark element, and changing the candidate The position of the vertex in its spherical coordinate system, making it
The value of θ is equal to the row index code value L and the column index code value J of the watermark element, respectively, and the r value thereof is equal to the element value I of the watermark element to embed the index code value and the energy value of the watermark element; The candidate vertices are culled in all candidate vertices satisfying the screening condition, and all the watermark elements of the original watermark image are traversed until the above is completely embedded;

The blind watermark extraction detecting step S2 includes:

Step S21: Filter all possible embedded vertices satisfying the screening condition from the three-dimensional model to be detected according to the first preset screening condition;

Step S22, establishing a local spherical coordinate system under the custom embedding vertices, and determining a spherical coordinate value of each of the possible embedded vertices
among them,
θ, r are three attribute values of each possible embedding vertex that satisfy the screening condition;

Step S23, combining each of the three-dimensional models with the attribute values of the possible embedded vertices The possible embedding vertices of the screening condition are used for index value and energy value extraction to obtain a matrix S4;

Step S24, inversely transforming the matrix S4 back into the airspace to obtain an extracted watermark image;

Step S25: Calculate a correlation between the extracted watermark image and the original watermark image by using a watermark correlation calculation formula, and if the correlation is greater than an experience threshold, prove that the three-dimensional model includes a watermark.
The method of claim 3, wherein the step S11 comprises:

Step S111, using a two-dimensional discrete cosine transform formula to perform two-dimensional discrete cosine transform on the original watermark image to be embedded, to obtain a spectral matrix S1 of the discrete cosine transform domain of the original watermark image;

The two-dimensional discrete cosine transform formula is:

Where u=0,1,...,W1-1;v=0,1...,W2-1;

The size of the original watermark picture is W1×W2, W1 represents the length of the original watermark picture, W2 represents the width of the original watermark picture, and the matrix of the original watermark picture is represented as f(x, y), the original watermark The picture is denoted as S, and S1(u, v) represents a spectrum matrix obtained by discrete cosine transform of the matrix f(x, y) of the original watermark picture;

Step S112, extracting the positive and negative coefficient matrix of the spectrum matrix S1 as the recovery key K1, performing absolute value on the spectrum matrix S1, and performing normalization processing to obtain a matrix S2, and the matrix S2 The maximum value as the recovery key K2;

Step S113, performing Arnold scrambling on the matrix S2 by using a scrambling matrix to obtain a matrix S3, where the scrambling matrix is:

Where a and b are the set parameters, k is the number of iterations, W2 is the width of the original watermark picture, a, b, k, W2 are used as the recovery key K3; each element in the matrix S3 has three attributes Value (m, n, I), where m is the row index value, n is the column index value, and I is the element value, ie the energy value;

Step S114, using the following formula to respectively encode the attribute values m, n of each element in the matrix S3 into an angle value, to obtain a new attribute value (L, J, I) of each element of the matrix S3;

Where α and β are preset constants, α is a value between 0 and 0.5, β is a value between 0 and 1, L is the row index code value, J is the column index code value, and I is The energy value, L i represents the index code value of the i-th row, and J j represents the index code value of the j-th column.
The method for detecting and embedding a three-dimensional blind watermark according to claim 1, wherein the first preset screening condition comprises: a first screening condition, a second screening condition, and a third screening condition, wherein the first A screening condition is that the maximum ratio of the distance between the one-ring neighbor point of the vertex and the centroid of the one-ring neighbor point needs to be smaller than the threshold A, and the second screening condition is that the normal angle of the two different definitions of the centroid of the neighboring point is required. Greater than the threshold B, the third screening condition is a non-neighboring point between the candidate vertices;

The step S12 specifically includes:

Step S121, the centroid link neighbor point three-dimensional model is calculated to be embedded watermarks vertices V i to Z i, are connected to a ring neighbor point Z i and V i, and to calculate the appropriate distance, if each of the distance If the maximum ratio is greater than the threshold A, the vertex is culled, and all the vertices of the three-dimensional model are traversed by the operation to obtain all the first candidate vertices satisfying the condition;

Step S122, calculating normal directions n a and n t for the first candidate vertex, if the angle between the normal n a and the normal n t is less than the threshold B, then the first candidate vertex is eliminated, for all The first candidate vertex traverses this operation to obtain all second candidate vertices satisfying the condition;

Wherein the normal n a is defined as:

Connect a circle of neighbor points of the centroid point Z i and V i to obtain a number of new triangles with the centroid point as the apex, denoted as T(Z i ), and the number of triangles is denoted as N, the new triangle The normal is denoted as n j , and the mean of all the normals of the triangle is taken as the normal n a of the centroid point Z i , namely:

The normal n t is defined as:

Calculating another normal n t of the centroid according to the square of the normal n j and the area S j of each of the triangles, namely:

Step S123, it is checked whether all the second candidate vertices satisfying the condition have vertices which are mutually neighbors, and if so, cull them, and the remaining second candidate vertices are candidate vertices of the watermark embedding.
The method for detecting and embedding a three-dimensional blind watermark according to claim 1, wherein in the step S13, the spherical coordinate value
Finding the spherical coordinate value of the candidate vertex for the customized encoded spherical coordinate value
The formula is as follows:

r=sqrt((xx 0 ) 2 +(yy 0 ) 2 +(zz 0 ) 2 )/ρ

θ=arccos(|zz 0 |/r)

Where (x, y, z) is the coordinate value of the candidate vertex in the Cartesian coordinate system, (x 0 , y 0 , z 0 ) is the centroid point coordinate of the ring neighbor of the candidate vertex, and ρ is the centroid point to the neighbor point The mean of the distance.
The method of claim 3, wherein the step S21 comprises:

Step S211, the centroid point of a neighboring ring vertex V i of the three-dimensional model calculated Z i to be detected, are connected to a ring neighbor point Z i and V i, and to calculate the appropriate distance, if the respective distance If the maximum ratio is greater than the threshold A, the vertex is culled, and all the vertices of the three-dimensional model are traversed by the operation to obtain all the first possible embedded vertices satisfying the condition;

Step S212, calculating normal directions n a and n t for the first possible embedding vertex, and if the angle between the normal n a and the normal n t is less than the threshold B, the first possible embedding vertex is eliminated. Traversing this operation for all first possible embedding vertices, yielding all second possible embedding vertices that satisfy the condition;

Wherein the normal n a is defined as:

Connect the centroid neighbor points of the centroid point Z i and Vi to obtain a number of new triangles with the centroid point as the apex, denoted as T(Z i ), and the number of triangles is denoted by N, the method of the new triangle The direction is denoted as n j , and the mean of all the normals of the triangle is taken as the normal n a of the centroid point Z i , namely:

The normal n t is defined as:

Calculating another normal n t of the centroid according to the normal n j and the area S j of each of the triangles, namely:

Step S213, it is checked whether all the second possible embedding vertices satisfying the condition have vertices which are mutually neighbors, and if so, cull them, and the remaining second possible embedding vertices are the finally selected possible embedding vertices.
The method of claim 3, wherein the step S23 comprises:

Step S231, using the following formula to attribute the attribute value of the possible embedding vertex
Perform anti-coding;

Step S232, establishing a zero matrix S4 of the same size as the matrix S3, each matrix element has three attributes (m, n, 0), m and n are row and column values, respectively, and the possible vertices are embedded.
When the value of θ satisfies the following inequality, the r value of the possible embedding vertex is replaced by the zero value of the m row and n column of the matrix S3; all possible embedding vertices are traversed to find the vertex satisfying the index threshold condition, and the r value is replaced Corresponding to the zero value of the matrix S4 of the index, a new matrix S4 is obtained;

Where λ1 and λ2 respectively represent a set threshold value, which satisfies the above inequality
The value of θ is the index value, and the corresponding r value is the energy value.
The method of claim 3, wherein the step S24 comprises:

Step S241, using the recovery key K3 as a restore key, the Arnold reduction of the matrix S4, to obtain a reduction matrix S5;

Step S242, multiplying the elements in the reduction matrix S5 by the recovery key K2 to obtain a matrix S6;

Step S243, using the recovery key K1 to multiply the matrix S6, thereby obtaining a spectrum matrix S7;

Step S244, performing a two-dimensional inverse discrete cosine transform on the spectrum matrix S7 by using a two-dimensional inverse discrete cosine transform formula to obtain a matrix S8, where the matrix S8 is an extracted watermark image, and the two-dimensional inverse discrete cosine transform formula as follows:

Where u=0,1,...,W1-1;v=0,1...,W2-1;

Wherein, S8(x, y) represents a watermark image matrix S8 obtained by inverse discrete cosine transform from the extracted spectral matrix S7(u, v).
The method for detecting and embedding a three-dimensional blind watermark according to claim 2, wherein in the step S25, the watermark correlation calculation formula is:

Where W1 represents the length of the original watermark picture, W2 represents the width of the original watermark picture, S(i,j) represents the matrix of the original watermark picture, S(i,j) represents the matrix of the extracted watermark picture, and σ represents the watermark correlation degree.
A device for embedding and extracting a three-dimensional blind watermark in a partial spherical coordinate system, wherein the device comprises: a blind watermark embedding module and a blind watermark extraction detecting module;

The blind watermark embedding module includes:

a pre-processing sub-module, configured to pre-process the original watermark image to be embedded, to obtain an attribute value (L, J, I) of the watermark element of the original watermark image;

Where L is the row index code value, J is the column index code value, and I is the element value, ie the energy value;

a first screening sub-module, configured to filter, from the three-dimensional model of the watermark to be embedded, all candidate vertices satisfying the screening condition according to the first preset screening condition;

a first coordinate system establishing submodule, configured to establish a local spherical coordinate system under the custom vertices, and determine a spherical coordinate value of each of the candidate vertices
among them,
θ, r are three attribute values of each candidate vertex satisfying the screening condition;

a watermark embedding sub-module, configured to compare a property value of the watermark element with an attribute value of all candidate vertices satisfying a filter condition, and search for a candidate vertex that has the smallest difference from the attribute value of the watermark element as a point at which the watermark element is embedded , changing the position of this candidate vertex in its spherical coordinate system, making it
The value of θ is equal to the row index code value L and the column index code value J of the watermark element, respectively, and the r value thereof is equal to the element value I of the watermark element to embed the index code value and the energy value of the watermark element; The candidate vertices are culled in all candidate vertices satisfying the screening condition, and all the watermark elements of the original watermark image are traversed until the above is completely embedded;

The blind watermark extraction detection module includes:

a second screening sub-module, configured to filter, from the three-dimensional model to be detected, all possible embedded vertices satisfying the screening condition according to the first preset screening condition;

a second coordinate system establishing submodule, configured to establish a local spherical coordinate system under the custom embedding vertices, and determine a spherical coordinate value of each of the possible embedded vertices
among them,
θ, r are three attribute values of each possible embedding vertex that satisfy the screening condition;

An extraction submodule, configured to perform an index value and an energy value extraction on the possible embedded vertex satisfying the filter condition in the three-dimensional model in combination with the attribute value of the possible embedding vertex, to obtain a matrix S4;

An inverse transform sub-module, configured to inverse transform the matrix S4 back into the airspace to obtain an extracted watermark image;

The determining sub-module is configured to calculate a correlation between the extracted watermark image and the original watermark image by using a watermark correlation calculation formula, and if the correlation is greater than an experience threshold, prove that the three-dimensional model includes a watermark.