CN108470176B - Stereo image visual saliency extraction method based on frequency domain sparse representation - Google Patents
Stereo image visual saliency extraction method based on frequency domain sparse representation Download PDFInfo
- Publication number
- CN108470176B CN108470176B CN201810066806.9A CN201810066806A CN108470176B CN 108470176 B CN108470176 B CN 108470176B CN 201810066806 A CN201810066806 A CN 201810066806A CN 108470176 B CN108470176 B CN 108470176B
- Authority
- CN
- China
- Prior art keywords
- value
- image
- matrix
- column
- row
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/513—Sparse representations
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a three-dimensional image visual saliency extraction method based on frequency domain sparse representation, which is used for acquiring a small-size three-channel image and a small-size normalized disparity map of a left viewpoint image of a tested three-dimensional image; blocking the three-channel image and the normalized disparity map to obtain a matrix corresponding to a blocked image; forming a quaternion matrix by the four matrixes, obtaining a transformed quaternion matrix after two-dimensional quaternion Fourier transform, and obtaining a low frequency component diagram according to low frequency components of a real part and three imaginary parts; extracting a sparse weight matrix of the low frequency component diagram by using a sparse representation dictionary; acquiring two central periphery significant images according to the sparse weight matrix and the central preference image; blurring the two images with significant centers around, and fusing to obtain a fused image; reinforcing the periphery of the center of the fused image by using the center preference image; enhancing the size conversion of the image around the center to obtain a visual salient image; has the advantages of stronger extraction stability and higher extraction accuracy.
Description
Technical Field
The invention relates to a processing method of image signals, in particular to a method for extracting visual saliency of a three-dimensional image based on frequency domain sparse representation.
Background
After receiving the natural image, people need to distinguish information resources of different levels, so that when processing natural image information, people can perform grading processing on different information, and thus, the selected characteristics are shown. People are not evenly focused on various areas of an image when watching the image or video clip, but rather preferentially process the more interesting part of the semantic information. Computing image salient regions is an important research area in the field of computer vision and content-based video detection. With the rapid development of stereoscopic image projection and acquisition equipment, the visual saliency detection of stereoscopic images also becomes a very important research content.
The stereoscopic image is not a simple expansion of a plane image, and the process of generating the stereoscopic image by sensing the stereoscopic image by human eyes is not a simple superposition process of a left viewpoint image and a right viewpoint image, so that the stereoscopic vision characteristic is not a simple expansion of a plane vision characteristic. However, the current stereo image visual saliency extraction method still stays on the plane image visual saliency extraction method. Therefore, how to effectively extract stereoscopic features from a stereoscopic image and how to make the extracted stereoscopic features conform to the observation habit of the human visual system are all problems that need to be studied in the process of extracting a visual saliency map of the stereoscopic image.
Disclosure of Invention
The invention aims to provide a frequency domain sparse representation-based stereo image visual saliency extraction method which accords with saliency semantic features and has stronger extraction stability and higher extraction accuracy.
The technical scheme adopted by the invention for solving the technical problems is as follows: a stereoscopic image visual saliency extraction method based on frequency domain sparse representation is characterized by comprising the following steps:
① for any one of the test stereo images StestWill StestIs transformed into a Lab color space and the size is transformed into a size of 200 × 200 pixels, and the transformed image is denoted as { L }Lab200(x1,y1) }; then will { LLab200(x1,y1) The L channel image, the a channel image and the b channel image of the image are correspondingly recorded as { L }Lab200,L(x1,y1)}、{LLab200,a(x1,y1)}、{LLab200,b(x1,y1) }; wherein S istestHas a width of W, StestHas a height of H, 1 is less than or equal to x1≤200,1≤y1≤200,LLab200(x1,y1) Represents { LLab200(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,L(x1,y1) Represents { LLab200,L(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,a(x1,y1) Represents { LLab200,a(x1,y1) The position of the middle coordinate is (x)1,y1) Is/are as followsPixel value, L, of a pixel pointLab200,b(x1,y1) Represents { LLab200,b(x1,y1) The position of the middle coordinate is (x)1,y1) The pixel value of the pixel point of (1);
will StestThe parallax map of (2) is scaled to a size of 200 × 200 pixels, and the scaled image is denoted as { D }200(x1,y1) }; then will { D200(x1,y1) Normalizing the pixel value of each pixel point in the pixel to [0, 1 ]]Numerical value range, and the normalized image is recorded as { D0,1(x1,y1) }; wherein D is200(x1,y1) Represents { D200(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (2), D0,1(x1,y1) Represents { D0,1(x1,y1) The position of the middle coordinate is (x)1,y1) The pixel value of the pixel point of (1);
② will { LLab200,L(x1,y1)}、{LLab200,a(x1,y1)}、{LLab200,b(x1,y1)}、{D0,1(x1,y1) Are divided intoNon-overlapping image blocks of size 8 × 8, and LLab200,L(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { LLab200,L(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; and will be { LLab200,a(x1,y1) Divide the block image to get the corresponding matrix of the block image as From { LLab200,a(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; will { LLab200,b(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { LLab200,b(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; will { D0,1(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { D0,1(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; wherein m is more than or equal to 11≤8,1≤n1≤8, Represents { LLab200,L(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),represents { LLab200,a(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),represents { LLab200,b(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),represents { D0,1(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element of (c);
③ will beForm a quaternion matrix, which is denoted asWill be provided withThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The quaternion value of the element(s) is noted Wherein i, j and k are all imaginary number units;
④ pairsPerforming two-dimensional quaternion Fourier transform on each sub-matrix to obtainThe transformed quaternion matrix of (2), noteWill be provided withThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The quaternion value of the element(s) is noted Then will beThe real part, i-imaginary part, j-imaginary part and k-imaginary part of (A) are correspondingly marked as Followed by extraction Respective low frequency components, corresponding to Wherein QFT () represents a two-dimensional quaternion Fourier transform function,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (1) m2≤4,1≤n2≤4,To representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element of (c);
⑤ are in accordance with ObtainingCorresponding low frequency component diagram, noteWill be provided withThe coordinate position in the 8 × 8 area of the u-th row and the v-th column in (m)1,n1) The pixel value of the pixel point is recorded asWherein the content of the first and second substances,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1-4,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1-4) of the value of the element,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1-4,n1-4) value of an element;
⑥ by the Lasso method andfor sparse representation of dictionaries, extractionThe sparse weight of each 8 × 8 region in the image is obtainedCorresponding sparse weight matrix, denotedWherein, in the Lasso method, a weighting parameter lambda is adopted to select a Lagrange penalty term, u is more than or equal to 11≤25,1≤v1≤25,Is shown inRepresenting dictionary pairs for sparsenessU-th row and v-th column in (1) 8 × 8 area generated when sparse representation is performed1Line v1The weight of the column has a subscript of (m)1,n1) The value of the element(s) of (c), to representThe sub-matrix of the u-th row and the v-th column in (1) has a subscript of (u)1,v1) The value of the element of (c);
⑦ obtains an absolute value sum image of 25 × 25 pixel sizes, noted as { W }abs(u, v) }, will { WabsThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as Wabs(u,v),And a center preference image of 25 × 25 pixel size is acquired, denoted as { C }saliency25(u1,v1) Will { C }saliency25(u1,v1) The position of the coordinate in the Chinese character is (u)1,v1) The pixel value of the pixel point is marked as Csaliency25(u1,v1),A first center four-around salient image of 25 × 25 pixel size is then acquired, denoted as Ccs(u, v) }, will { CcsThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as Ccs(u,v), And a second center-surrounding salient image of 25 × 25 pixel size is acquired, denoted as { Cnear(u, v) }, will { Cnear(u, v) } the pixel value of the pixel point with the coordinate position of (u, v) is recorded as Then using Gaussian filter template pair { Ccs(u, v) } fuzzy processing is performed, and { C is usedcs(u, v) } blurring processing the obtained image is marked as { Ccs-g(u, v) }; and using a Gaussian filter template pair { Cnear(u, v) } fuzzy processing is performed, and { C is usednear(u, v) } blurring processing the obtained image is marked as { Cnear-g(u, v) }; wherein abs () is the function of taking the absolute value, e is the natural base, u0Represents { Csaliency25(u1,v1) Abscissa of center pixel point, v0Represents { Csaliency25(u1,v1) Ordinate, δ, of the central pixel point ofFDenotes a first center preference parameter, δDDenotes a second center preference parameter, Ccs-g(u, v) represents { Ccs-g(u, v) } the pixel value of the pixel point with the coordinate position of (u, v), Cnear-g(u, v) represents { Cnear-gThe pixel value of a pixel point with (u, v) as the coordinate position in (u, v) }, and the parameter for controlling the fuzzy degree is taken when the Gaussian filtering template is appliedIs 5, and the filter width control parameter X is takenGIs 5, and the control filter height parameter Y is takenGA value of 5;
⑧ pairs { Ccs-g(u, v) } and { Cnear-g(u, v) } to obtain a fused image, which is marked as { CF(u, v) }, will { CFThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as CF(u,v),CF(u,v)=0.3×Cnear-g(u,v)+0.7×Ccs-g(u, v); then use { Csaliency25(u1,v1) Are { C } pairF(u, v) } center-around enhancement, resulting in a center-around enhanced image of 25 × 25 pixel size, denoted as { CFO(u, v) }, will { CFOThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as CFO(u,v),CFO(u,v)=CF(u,v)×Csaliency25(u1,v1) (ii) a Then pair { CFO(u, v) } image size conversion to obtain StestIs marked as { SF(x, y) }; wherein S isF(x, y) represents SFThe coordinate position in (x, y) is the pixel value of the pixel point of (x, y), x is more than or equal to 1 and less than or equal to W, and y is more than or equal to 1 and less than or equal to H.
In the step ①, the process is described,where min () is the minimum function and max () is the maximum functionA function.
In the step ④, the process is described,
to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representRow u and column v of (1)The subscript in the matrix is (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2+4) element value.
Compared with the prior art, the invention has the advantages that:
1) the method utilizes the multi-channel image and the disparity map, carries out block processing to obtain the quaternion matrix, then carries out two-dimensional quaternion Fourier transform on the quaternion matrix, and utilizes the transformed quaternion matrix to carry out subsequent processing to be beneficial to parallel calculation, thereby being in line with the observation habit of the human visual system and ensuring that the method of the invention is in line with the obvious semantic features.
2) The method of the invention fuses the characteristics of the depth information into the multi-channel image corresponding to the left viewpoint image to obtain the quaternion matrix, thus effectively improving the accuracy and stability of the detection of the visual saliency of the stereo image.
3) The method of the invention utilizes the low-frequency component of the two-dimensional quaternion Fourier transform to carry out sparse representation and uses less elements to carry out sparse representation, thereby improving the running speed of the method of the invention.
4) The method of the invention combines various preferences by using a sparse representation mode, improves the accuracy of the visual saliency detection of the stereo image and better plays the role of various preferences.
Drawings
Fig. 1 is a block diagram of the overall implementation of the method of the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying examples.
The invention provides a frequency domain sparse representation-based stereo image visual saliency extraction method, the overall implementation block diagram of which is shown in FIG. 1, and the method comprises the following steps:
① pairsTesting the stereo image S on any onetestWill StestIs transformed into a Lab color space and the size is transformed into a size of 200 × 200 pixels, and the transformed image is denoted as { L }Lab200(x1,y1) }; then will { LLab200(x1,y1) The L channel image, the a channel image and the b channel image of the image are correspondingly recorded as { L }Lab200,L(x1,y1)}、{LLab200,a(x1,y1)}、{LLab200,b(x1,y1) }; wherein S istestHas a width of W, StestHas a height of H, 1 is less than or equal to x1≤200,1≤y1≤200,LLab200(x1,y1) Represents { LLab200(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,L(x1,y1) Represents { LLab200,L(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,a(x1,y1) Represents { LLab200,a(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,b(x1,y1) Represents { LLab200,b(x1,y1) The position of the middle coordinate is (x)1,y1) The pixel value of the pixel point of (1).
Will StestThe parallax map of (2) is scaled to a size of 200 × 200 pixels, and the scaled image is denoted as { D }200(x1,y1) }; then will { D200(x1,y1) Normalizing the pixel value of each pixel point in the pixel to [0, 1 ]]Numerical value range, and the normalized image is recorded as { D0,1(x1,y1) }; wherein D is200(x1,y1) Represents { D200(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (2), D0,1(x1,y1) Represents { D0,1(x1,y1) The position of the middle coordinate is (x)1,y1) The pixel value of the pixel point of (1).
In this embodiment, at step ①,wherein, min () is a function of taking the minimum value, and max () is a function of taking the maximum value.
② will { LLab200,L(x1,y1)}、{LLab200,a(x1,y1)}、{LLab200,b(x1,y1)}、{D0,1(x1,y1) Are divided intoNon-overlapping image blocks of size 8 × 8, and LLab200,L(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { LLab200,L(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; and will be { LLab200,a(x1,y1) Divide the block image to get the corresponding matrix of the block image as From { LLab200,a(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; will { LLab200,b(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { LLab200,b(x1,y1) Dividing image blocks to obtain all image block pairs in block imagesConstructing a corresponding sub-matrix; will { D0,1(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { D0,1(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; wherein m is more than or equal to 11≤8,1≤n1≤8, Represents { LLab200,L(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (b) { L }Lab200,L(x1,y1) The submatrix corresponding to the image block of the nth row and the vth column in the block image obtained by dividing the image block isRow (u) and column (v) of (1),represents { LLab200,a(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (b) { L }Lab200,a(x1,y1) The submatrix corresponding to the image block of the nth row and the vth column in the block image obtained by dividing the image block isRow (u) and column (v) of (1),represents { LLab200,b(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (b) { L }Lab200,b(x1,y1) The submatrix corresponding to the image block of the nth row and the vth column in the block image obtained by dividing the image block isRow (u) and column (v) of (1),represents { D0,1(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (b) { D }0,1(x1,y1) The submatrix corresponding to the image block of the nth row and the vth column in the block image obtained by dividing the image block isRow (u) and column (v).
③ will beForm a quaternion matrix, which is denoted asWill be provided withThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The quaternion value of the element(s) is noted Wherein i, j and k are all imaginary number units.
④ pairsEach sub-matrix in the array is subjected to the existing two-dimensional quaternion Fourier transform to obtainThe transformed quaternion matrix of (2), noteWill be provided withThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The quaternion value of the element(s) is noted Then will beThe real part, i-imaginary part, j-imaginary part and k-imaginary part of (A) are correspondingly marked as Followed by extraction Respective low frequency components, corresponding to Wherein QFT () represents an existing two-dimensional quaternion Fourier transform function,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (1) m2≤4,1≤n2≤4,To representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s).
In this embodiment, at step ④, wherein the content of the first and second substances,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2+4) element value.
⑤ are in accordance with ObtainingCorresponding low frequency component diagram, noteWill be provided withThe coordinate position in the 8 × 8 area of the u-th row and the v-th column in (m)1,n1) The pixel value of the pixel point is recorded asWherein the content of the first and second substances,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1-4,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1-4) of the value of the element,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1-4,n1-4) value of an element.
⑥ use the existing Lasso method andfor sparse representation of dictionaries, extractionThe sparse weight of each 8 × 8 region in the image is obtainedCorresponding sparse weight matrix, denotedIn the Lasso method, a weighting parameter λ is used to select a lagrangian penalty term, where λ is 0.4 and u is 1 ≦ u in this embodiment1≤25,1≤v1≤25,Is shown inRepresenting dictionary pairs for sparsenessU-th row and v-th column in (1) 8 × 8 area generated when sparse representation is performed1Line v1The weight of the column has a subscript of (m)1,n1) The value of the element(s) of (c), to representThe sub-matrix of the u-th row and the v-th column in (1) has a subscript of (u)1,v1) The value of the element(s).
⑦ obtains an absolute value sum image of 25 × 25 pixel sizes, noted as { W }abs(u, v) }, will { WabsThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as Wabs(u,v),And a center preference image of 25 × 25 pixel size is acquired, denoted as { C }saliency25(u1,v1) Will { C }saliency25(u1,v1) The position of the coordinate in the Chinese character is (u)1,v1) The pixel value of the pixel point is marked as Csaliency25(u1,v1),A first center four-around salient image of 25 × 25 pixel size is then acquired, denoted as Ccs(u, v) }, will { CcsThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as Ccs(u,v), And a second center-surrounding salient image of 25 × 25 pixel size is acquired, denoted as { Cnear(u, v) }, will { Cnear(u, v) } the pixel value of the pixel point with the coordinate position of (u, v) is recorded as Then using the existing Gaussian filter template pair { Ccs(u, v) } fuzzy processing is performed, and { C is usedcs(u, v) } blurring processing the obtained image is marked as { Ccs-g(u, v) }; and using the existing Gaussian filter template pair { Cnear(u, v) } fuzzy processing is performed, and { C is usednear(u, v) } blurring processing the obtained image is marked as { Cnear-g(u, v) }; wherein abs () is an absolute value function, e is a natural base, e is 2.7182818284 …, u0Represents { Csaliency25(u1,v1) Abscissa of center pixel point, v0Represents { Csaliency25(u1,v1) Ordinate, δ, of the central pixel point ofFIndicates a first center preference parameter, u in this example0=12.5、v0=12.5、δF=114,δDRepresents a second center preference parameter, in this example, taken as δD=114,Ccs-g(u, v) represents { Ccs-g(u, v) } the pixel value of the pixel point with the coordinate position of (u, v), Cnear-g(u, v) represents { Cnear-gThe pixel value of a pixel point with (u, v) as the coordinate position in (u, v) }, and the parameter for controlling the fuzzy degree is taken when the Gaussian filtering template is appliedIs 5, and the filter width control parameter X is takenGIs 5, and the control filter height parameter Y is takenGThe value of (A) is 5.
⑧ pairs { Ccs-g(u, v) } and { Cnear-g(u, v) } carrying outFusing to obtain a fused image, and marking as { CF(u, v) }, will { CFThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as CF(u,v),CF(u,v)=0.3×Cnear-g(u,v)+0.7×Ccs-g(u, v); then use { Csaliency25(u1,v1) Are { C } pairF(u, v) } center-around enhancement, resulting in a center-around enhanced image of 25 × 25 pixel size, denoted as { CFO(u, v) }, will { CFOThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as CFO(u,v),CFO(u,v)=CF(u,v)×Csaliency25(u1,v1) (ii) a Then pair { CFO(u, v) } image size conversion to obtain StestIs marked as { SF(x, y) }; wherein S isF(x, y) represents SFThe coordinate position in (x, y) is the pixel value of the pixel point of (x, y), x is more than or equal to 1 and less than or equal to W, and y is more than or equal to 1 and less than or equal to H.
To verify the feasibility and effectiveness of the method of the invention, experiments were performed.
Here, a three-dimensional human eye tracking database (3Deye-tracking database) provided by south university of france was used to analyze the accuracy and stability of the method of the present invention. Here, 3 common objective parameters for evaluating the visually significant extraction method are used as evaluation indexes, i.e., Pearson correlation coefficient (PLCC), Kullback-Leibler divergence coefficient (KLD), AUC parameter (AUC) of the Area Under the receiving and operating characteristics.
The method is used for obtaining the visual salient image of each three-dimensional image in the three-dimensional human eye tracking database provided by southern university of France, and comparing the visual salient image with the subjective visual salient image (existing in the three-dimensional human eye tracking database) of each three-dimensional image in the three-dimensional human eye tracking database, wherein the higher the PLCC and AUC values and the lower the KLD value, the better the consistency between the visual salient image extracted by the method and the subjective visual salient image is. The relevant indices of PLCC, KLD and AUC that reflect the visually significant extraction performance of the method of the invention are listed in table 1. As can be seen from the data listed in Table 1, the accuracy and stability of the visually significant image and the subjective visually significant image extracted by the method are very good, which indicates that the objective extraction result is more consistent with the result of subjective perception of human eyes, and is enough to illustrate the feasibility and effectiveness of the method.
TABLE 1 accuracy and stability of visually significant images and subjective visually significant images extracted by the method of the present invention
Claims (3)
1. A stereoscopic image visual saliency extraction method based on frequency domain sparse representation is characterized by comprising the following steps:
① for any one of the test stereo images StestWill StestIs transformed into a Lab color space and the size is transformed into a size of 200 × 200 pixels, and the transformed image is denoted as { L }Lab200(x1,y1) }; then will { LLab200(x1,y1) The L channel image, the a channel image and the b channel image of the image are correspondingly recorded as { L }Lab200,L(x1,y1)}、{LLab200,a(x1,y1)}、{LLab200,b(x1,y1) }; wherein S istestHas a width of W, StestHas a height of H, 1 is less than or equal to x1≤200,1≤y1≤200,LLab200(x1,y1) Represents { LLab200(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,L(x1,y1) Represents { LLab200,L(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,a(x1,y1) Represents { LLab200,a(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (1), LLab200,b(x1,y1) Represents { LLab200,b(x1,y1) The position of the middle coordinate is (x)1,y1) The pixel value of the pixel point of (1);
will StestThe parallax map of (2) is scaled to a size of 200 × 200 pixels, and the scaled image is denoted as { D }200(x1,y1) }; then will { D200(x1,y1) Normalizing the pixel value of each pixel point in the pixel to [0, 1 ]]Numerical value range, and the normalized image is recorded as { D0,1(x1,y1) }; wherein D is200(x1,y1) Represents { D200(x1,y1) The position of the middle coordinate is (x)1,y1) Pixel value of the pixel point of (2), D0,1(x1,y1) Represents { D0,1(x1,y1) The position of the middle coordinate is (x)1,y1) The pixel value of the pixel point of (1);
② will { LLab200,L(x1,y1)}、{LLab200,a(x1,y1)}、{LLab200,b(x1,y1)}、{D0,1(x1,y1) Are divided intoNon-overlapping image blocks of size 8 × 8, and LLab200,L(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { LLab200,L(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; and will be { LLab200,a(x1,y1) Divide the block image to get the corresponding matrix of the block image as From { LLab200,a(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; will { LLab200,b(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { LLab200,b(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; will { D0,1(x1,y1) Divide the block image to get the corresponding matrix of the block image asFrom { D0,1(x1,y1) Dividing image blocks to obtain sub-matrixes corresponding to all the image blocks in the block images; wherein m is more than or equal to 11≤8,1≤n1≤8, Represents { LLab200,L(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),represents { LLab200,a(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),represents { LLab200,b(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),represents { D0,1(x1,y1) Dividing the image blocks to obtain the image blocks, wherein the coordinate position in the image block of the u row and the v column in the block image is (m)1,n1) The pixel value of the pixel point of (a),also showsThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element of (c);
③ will beForm a quaternion matrix, which is denoted asWill be provided withThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The quaternion value of the element(s) is noted Wherein i, j and k are all imaginary number units;
④ pairsPerforming two-dimensional quaternion Fourier transform on each sub-matrix to obtainThe transformed quaternion matrix of (2), noteWill be provided withThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The quaternion value of the element(s) is noted Then will beThe real part, i-imaginary part, j-imaginary part and k-imaginary part of (A) are correspondingly marked as Followed by extraction Respective low frequency components, corresponding to Wherein QFT () represents a two-dimensional quaternion Fourier transform function,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) Value of (1) m2≤4,1≤n2≤4,To representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element of (c);
⑤ are in accordance with ObtainingCorresponding low frequency component diagram, noteWill be provided withThe coordinate position in the 8 × 8 area of the u-th row and the v-th column in (m)1,n1) The pixel value of the pixel point is recorded as Wherein the content of the first and second substances,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1-4,n1) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1,n1-4) of the value of the element,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)1-4,n1-4) value of an element;
⑥ by the Lasso method andfor sparse representation of dictionaries, extractionThe sparse weight of each 8 × 8 region in the image is obtainedCorresponding sparse weight matrix, denotedWherein, in the Lasso method, a weighting parameter lambda is adopted to select a Lagrange penalty term, u is more than or equal to 11≤25,1≤v1≤25,Is shown inRepresenting dictionary pairs for sparsenessU-th row and v-th column in (1) 8 × 8 area generated when sparse representation is performed1Line v1The weight of the column has a subscript of (m)1,n1) The value of the element(s) of (c), to representThe sub-matrix of the u-th row and the v-th column in (1) has a subscript of (u)1,v1) The value of the element of (c);
⑦ obtains an absolute value sum image of 25 × 25 pixel sizes, noted as { W }abs(u, v) }, will { WabsThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as Wabs(u,v),And a center preference image of 25 × 25 pixel size is acquired, denoted as { C }saliency25(u1,v1) Will { C }saliency25(u1,v1) The position of the coordinate in the Chinese character is (u)1,v1) The pixel value of the pixel point is marked as Csaliency25(u1,v1),A first center four-around salient image of 25 × 25 pixel size is then acquired, denoted as Ccs(u, v) }, will { CcsThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as Ccs(u,v), And a second center-surrounding salient image of 25 × 25 pixel size is acquired, denoted as { Cnear(u, v) }, will { Cnear(u, v) } the pixel value of the pixel point with the coordinate position of (u, v) is marked as Cnear(u,v), Then using Gaussian filter template pair { Ccs(u, v) } fuzzy processing is performed, and { C is usedcs(u, v) } blurring processing the obtained image is marked as { Ccs-g(u, v) }; and using a Gaussian filter template pair { Cnear(u, v) } fuzzy processing is performed, and { C is usednear(u, v) } blurring processing the obtained image is marked as { Cnear-g(u, v) }; wherein abs () is the function of taking the absolute value, e is the natural base, u0Represents { Csaliency25(u1,v1) Abscissa of center pixel point, v0Represents { Csaliency25(u1,v1) Ordinate, δ, of the central pixel point ofFDenotes a first center preference parameter, δDDenotes a second center preference parameter, Ccs-g(u, v) represents { Ccs-g(u, v) } the pixel value of the pixel point with the coordinate position of (u, v), Cnear-g(u, v) represents { Cnear-gThe pixel value of a pixel point with (u, v) as the coordinate position in (u, v) }, and the parameter for controlling the fuzzy degree is taken when the Gaussian filtering template is appliedIs 5, and the filter width control parameter X is takenGIs 5, and the control filter height parameter Y is takenGA value of 5;
⑧ pairs { Ccs-g(u, v) } and { Cnear-g(u, v) } fusionThe resultant fused image is recorded as { CF(u, v) }, will { CFThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as CF(u,v),CF(u,v)=0.3×Cnear-g(u,v)+0.7×Ccs-g(u, v); then use { Csaliency25(u1,v1) Are { C } pairF(u, v) } center-around enhancement, resulting in a center-around enhanced image of 25 × 25 pixel size, denoted as { CFO(u, v) }, will { CFOThe pixel value of the pixel point with the coordinate position (u, v) in (u, v) is marked as CFO(u,v),CFO(u,v)=CF(u,v)×Csaliency25(u1,v1) (ii) a Then pair { CFO(u, v) } image size conversion to obtain StestIs marked as { SF(x, y) }; wherein S isF(x, y) represents SFThe coordinate position in (x, y) is the pixel value of the pixel point of (x, y), x is more than or equal to 1 and less than or equal to W, and y is more than or equal to 1 and less than or equal to H.
3. The method for extracting visual saliency of stereoscopic images based on frequency domain sparse representation according to claim 1 or 2, wherein in said step ④,
wherein the content of the first and second substances,to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representIn the sub-matrix of the u-th row and the v-th column of (1)Is marked by (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2,n2A value of the element of +4),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2) The value of the element(s) of (c),to representThe sub-matrix of the u-th row and the v-th column in (1) has a middle index of (m)2+4,n2+4) element value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810066806.9A CN108470176B (en) | 2018-01-24 | 2018-01-24 | Stereo image visual saliency extraction method based on frequency domain sparse representation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810066806.9A CN108470176B (en) | 2018-01-24 | 2018-01-24 | Stereo image visual saliency extraction method based on frequency domain sparse representation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108470176A CN108470176A (en) | 2018-08-31 |
CN108470176B true CN108470176B (en) | 2020-06-26 |
Family
ID=63266107
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810066806.9A Active CN108470176B (en) | 2018-01-24 | 2018-01-24 | Stereo image visual saliency extraction method based on frequency domain sparse representation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108470176B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112182492B (en) * | 2020-09-23 | 2022-12-16 | 广东工业大学 | Signal sparse representation method and device based on discrete quaternion Fourier transform |
CN112381838B (en) * | 2020-11-14 | 2022-04-19 | 四川大学华西医院 | Automatic image cutting method for digital pathological section image |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20130068182A (en) * | 2011-12-14 | 2013-06-26 | 한국전자통신연구원 | Methods of extracting saliency region and apparatuses for using the same |
CN104754320A (en) * | 2015-03-27 | 2015-07-01 | 同济大学 | Method for calculating 3D-JND threshold value |
CN105740859A (en) * | 2016-01-27 | 2016-07-06 | 电子科技大学 | Three-dimensional interest point detection method based on geometric measure and sparse optimization |
CN106682599A (en) * | 2016-12-15 | 2017-05-17 | 浙江科技学院 | Stereo image visual saliency extraction method based on sparse representation |
-
2018
- 2018-01-24 CN CN201810066806.9A patent/CN108470176B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20130068182A (en) * | 2011-12-14 | 2013-06-26 | 한국전자통신연구원 | Methods of extracting saliency region and apparatuses for using the same |
CN104754320A (en) * | 2015-03-27 | 2015-07-01 | 同济大学 | Method for calculating 3D-JND threshold value |
CN105740859A (en) * | 2016-01-27 | 2016-07-06 | 电子科技大学 | Three-dimensional interest point detection method based on geometric measure and sparse optimization |
CN106682599A (en) * | 2016-12-15 | 2017-05-17 | 浙江科技学院 | Stereo image visual saliency extraction method based on sparse representation |
Non-Patent Citations (5)
Title |
---|
A novel multiresolution spatiotemporal saliency detection;Guo C 等;《 IEEE Transactions on Image Processing》;20090825;第185-198页 * |
A Weighted Sparse Coding Framework for Saliency Detection;Nianyi Li 等;《The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)》;20151231;第5216-5223页 * |
Saliency Detection Using Quaternion Sparse Reconstruction;Zeng, Y 等;《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOP (ICCVW)》;20151218;第469-476页 * |
图像频域显著性检测;赵红苗;《中国优秀硕士学位论文全文数据库信息科技辑》;20160731;第I138-860页 * |
基于超复数傅里叶变换的自适应显著性检测;黄侃 等;《计算机应用》;20170615;第149-154页 * |
Also Published As
Publication number | Publication date |
---|---|
CN108470176A (en) | 2018-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107564025B (en) | Electric power equipment infrared image semantic segmentation method based on deep neural network | |
Li et al. | A weighted sparse coding framework for saliency detection | |
CN110353675B (en) | Electroencephalogram signal emotion recognition method and device based on picture generation | |
CN104834922B (en) | Gesture identification method based on hybrid neural networks | |
CN103020965B (en) | A kind of foreground segmentation method based on significance detection | |
CN106462771A (en) | 3D image significance detection method | |
CN108765414B (en) | No-reference stereo image quality evaluation method based on wavelet decomposition and natural scene statistics | |
CN106228528B (en) | A kind of multi-focus image fusing method based on decision diagram and rarefaction representation | |
CN103996195A (en) | Image saliency detection method | |
CN111754396B (en) | Face image processing method, device, computer equipment and storage medium | |
CN105320950A (en) | A video human face living body detection method | |
CN110827312B (en) | Learning method based on cooperative visual attention neural network | |
Liu et al. | Blind stereoscopic image quality assessment based on hierarchical learning | |
Zhou et al. | Blind 3D image quality assessment based on self-similarity of binocular features | |
CN105654142A (en) | Natural scene statistics-based non-reference stereo image quality evaluation method | |
CN111080670A (en) | Image extraction method, device, equipment and storage medium | |
CN108470176B (en) | Stereo image visual saliency extraction method based on frequency domain sparse representation | |
Karimi et al. | Blind stereo quality assessment based on learned features from binocular combined images | |
CN112669249A (en) | Infrared and visible light image fusion method combining improved NSCT (non-subsampled Contourlet transform) transformation and deep learning | |
Jiang et al. | Quality assessment for virtual reality technology based on real scene | |
CN107665488B (en) | Stereo image visual saliency extraction method | |
CN106682599B (en) | Sparse representation-based stereo image visual saliency extraction method | |
CN113705361A (en) | Method and device for detecting model in living body and electronic equipment | |
CN107886533B (en) | Method, device and equipment for detecting visual saliency of three-dimensional image and storage medium | |
CN115965844B (en) | Multi-focus image fusion method based on visual saliency priori knowledge |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220622 Address after: Unit a2203e, innovation Plaza, No. 2007, Pingshan Avenue, Liulian community, Pingshan street, Pingshan District, Shenzhen, Guangdong 518118 Patentee after: Shenzhen Muye Microelectronics Technology Co.,Ltd. Address before: 310023 No. 318 stay Road, Xihu District, Zhejiang, Hangzhou Patentee before: ZHEJIANG University OF SCIENCE AND TECHNOLOGY |