CN106251299A - A kind of high-efficient noise-reducing visual pattern reconstructing method - Google Patents
A kind of high-efficient noise-reducing visual pattern reconstructing method Download PDFInfo
- Publication number
- CN106251299A CN106251299A CN201610589708.4A CN201610589708A CN106251299A CN 106251299 A CN106251299 A CN 106251299A CN 201610589708 A CN201610589708 A CN 201610589708A CN 106251299 A CN106251299 A CN 106251299A
- Authority
- CN
- China
- Prior art keywords
- base
- topography
- window
- pixel point
- image
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000000007 visual effect Effects 0.000 title claims abstract description 40
- 238000012876 topography Methods 0.000 claims abstract description 75
- 238000012549 training Methods 0.000 claims abstract description 48
- 230000004936 stimulating effect Effects 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000006243 chemical reaction Methods 0.000 claims abstract description 8
- 210000004556 brain Anatomy 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 16
- 238000012216 screening Methods 0.000 claims description 6
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 238000013213 extrapolation Methods 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 238000002599 functional magnetic resonance imaging Methods 0.000 abstract description 15
- 238000005516 engineering process Methods 0.000 abstract description 8
- 238000012545 processing Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000003925 brain function Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000010365 information processing Effects 0.000 description 3
- 241000222065 Lycoperdon Species 0.000 description 2
- 241000768494 Polymorphum Species 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 230000000638 stimulation Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 210000001015 abdomen Anatomy 0.000 description 1
- 238000013398 bayesian method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 210000001320 hippocampus Anatomy 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000003973 paint Substances 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 210000001525 retina Anatomy 0.000 description 1
- 230000000392 somatic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Abstract
The invention discloses a kind of high-efficient noise-reducing visual pattern reconstructing method, belong to Biomedical Image mode identification technology.First the present invention carries out dimension-reduction treatment to characteristic vector, more corresponding topography based on different scale base carries out image conversion process to the contrast figure of the stimulating image of training sample, and builds the grader of correspondence;Simultaneously by the classification of training sample and prediction being obtained many groups the first prediction label, and then obtain the coefficient of colligation of each pixel of visual pattern;Input pending fMRI signal, obtain the second prediction label that different topographies base is corresponding, the coefficient of colligation of same pixel and the weighted sum of the second prediction label obtain the reconstruct label of current pixel point, thus obtain the visual pattern of reconstruct.The enforcement of the present invention not only can reconstruct visual pattern, and reduces reconstruct image noise.
Description
Technical field
This method belongs to Biomedical Image mode identification technology, is specifically related to the vision figure of functional MRI
As reconstructing method.
Background technology
Human brain has the advantages such as efficient, robust and anti-noise in complicated Vision information processing.Visual information is human perception
With one of topmost approach in the understanding external world.Brain has certain general character for different classes of natural scene processing,
Such as, the other region of temporo abdomen Hippocampus is the closely bound up brain domain that processes with scene information, but the processing to scene information
Also relying on multiple vision brain district, brain function characterization of visual information is high complexity.Up to now, brain is processed by we
Brain district and the encoding mechanism thereof of complicated natural scene classification are still known little about it.
In recent years, along with modern neuro image (brain electricity, brain magnetic, functional mri (fMRI), near-infrared optical brain merit
Energy imaging) and the development of computer technology, people have been gradually opened the gate understanding brain Vision information processing mechanism.Due to nothing
Wound property and the advantage of high spatial resolution, fMRI has played extremely important effect in Vision information processing, applies fMRI pair
Natural scene coding and decoding also achieve a series of important achievement.By fMRI technology, we are possible not only to detect volume
The brain district of code natural image, the process of tracking brain coding natural scene, it is also possible to enter according to corresponding cerebration signal
Row mode classification and visual information reconstruct (i.e. reappearing the natural image observed), and natural image and cerebration are entered
Row coupling modeling, and then be expected to cerebration is carried out real time information decoding.See Fig. 1, the overall think of of visual pattern reconstruct
Lu Shi: first, (flicker gridiron pattern image, if background is Lycoperdon polymorphum Vitt, pattern (geometry or letter) is black to allow tester watch stimulation figure
The stimulating images such as white flicker gridiron pattern image);Then, obtain flicker gridiron pattern image by magnetic resonance to stimulate in brain visual area
The fMRI signal of (as determined position, brain visual area by the experiment of retina Topological Mapping);Finally, mode identification method is used
This signal is carried out visual scene decoding.But, the brain function that it is critical only that extraction efficient stable of decoding brain function information is lived
Dynamic feature and the mode identification technology of correspondence.These difficult problems allow the research understanding human brain visual cognition movable become current undoubtedly
Most forward position, the direction most challenged in brain science research.
Visual pattern reconstruct is one of important research direction of visual information decoding.Trace back to eighties of last century the nineties,
Researcher is just had to use outer paint somatic nerves to put a signal to reconstruct visual scene, due to social production technology, the work in that age
The a series of restriction such as technology, although visual pattern reconstruct achieves certain effect, but reconstruction accuracy is also far from reaching
To preferable result, and subjects is had invasive damage.Along with rise and the development of mr techniques, the most non-invasive
Research brain become possibility, as use sparse polynomial logistic regression (SMLR) vision reconstructing method, although the method
Improve existing quality reconstruction, but the noise contained by reconstruction result is big.
Summary of the invention
The goal of the invention of the present invention is: for the problem of above-mentioned existence, it is provided that a kind of high-efficient noise-reducing visual pattern reconstruct
Method.
The high-efficient noise-reducing visual pattern reconstructing method of the present invention includes that training and vision reconstruct two parts, and it implements
Process is respectively
A. training step:
Step A1: input training sample, described training sample include the stimulating image FMRI signal in brain visual area, with
And with the contrast figure (the size normalization of contrast figure) of FMRI signal stimulating image one to one, wherein stimulating image is
Flicker checkerboard pattern;
Step A2: use N (N is more than 1) to plant topography base (φ1~φN) the contrast figure of stimulating image is converted
Process, obtain the N width changing image of same contrast figure:
Based on current topography base, respectively each pixel of stimulating image is carried out conversion process, take topography's base
Average contrast as the label of current pixel point, obtain the changing image of current topography base, wherein topography's base
Average contrast defined in the window of place for the ratio of number and chessboard sum of flicker chessboard.As follows in used
Nine kinds of topography's bases:
Topography base φ1: only include current pixel point 1 × 1 window;
Topography base φ2: include current pixel point abutment points right with it 1 × 2 window;
Topography base φ3: include current pixel point abutment points left with it 1 × 2 window;
Topography base φ4: include current pixel point with its on abutment points 2 × 1 window;;
Topography base φ5: include that current pixel point and its descend the window of the 2 × 1 of abutment points;
Topography base φ6: include current pixel point 2 × 2 window, wherein current pixel point is in the upper left of window
Angle;
Topography base φ7: include current pixel point 2 × 2 window, wherein current pixel point is in the upper right of window
Angle;
Topography base φ8: include current pixel point 2 × 2 window, wherein current pixel point is in the lower-left of window
Angle;
Topography base φ9: include current pixel point 2 × 2 window, wherein current pixel point is in the bottom right of window
Angle;
Step A3: the coefficient of colligation of the structure each pixel of visual pattern:
Step A3-1: each pixel position to the contrast figure of stimulating image, the respectively instruction of acquisition N kind topography base
Practice and predict label:
Training sample is divided into two subsets, and a subset is as training data, and a subset is as test data, not
Under same topography's base, based on changing image and FMRI signal, training data is carried out classifier training, obtain each pixel
At the first grader of different topographies base, wherein the classification of the first grader is the average contrast of each topography base
Value.Corresponding above-mentioned 9 kinds of topography's bases, wherein topography's base φ1Including two classes, all kinds of contrasts is respectively as follows: 0,1;
Topography base φ2、φ3、φ4、φ5Including three classes, all kinds of contrasts is respectively as follows: 0,0.5,1;Topography base φ6、φ7、
φ8、φ9Including five classes, all kinds of contrasts is respectively as follows: 0,0.25,0.5,0.75,1.
By the first grader, test data are carried out Classification and Identification, obtain each pixel position at N kind topography base
Under first prediction labelWherein i=1,2 ..., N;
Step A3-2: according to formulaIn multiple test data, the numerical value of residual epsilon will be made
(unsigned number word, or directly take residual epsilon absolute value or square) minimum ω1,ω2,...,ωNAs current pixel point
The coefficient of colligation of position, thus obtain the coefficient of colligation of each pixel of visual pattern, wherein CtrRepresent the contrast of stimulating image
The pixel value of figure;
Step A4: the value of average contrast based on N kind topography base divides classification, and training sample is carried out classification
Divide;And the characteristic vector of the FMRI signal of training sample is carried out Feature Selection, take the spy that front K classification judgement ability is maximum
Levying vector as Feature Selection result, wherein K is preset value, as used multi-class F-score feature selection mode, thus realizes
Dimension-reduction treatment to characteristic vector;
Under different topography's bases, carry out classifier training based on the FMRI signal after changing image and screening,
To pixel at the second grader of different topographies base, wherein the classification of the second grader is the flat of N kind topography base
All the value of contrast, above-mentioned 9 kinds of topography's bases, can be divided into five classes by training sample, and the contrast of all kinds of correspondences is respectively
For: 0,0.25,0.5,0.75,1.
Step B: visual pattern reconstructs
Step B1: input FMRI signal to be reconstructed, and the selection result based on step A4 treats the FMRI signal of reconstruct
Feature screen, by K characteristic vector after screening as the input of the second grader;
Step B2: recognition result based on the second grader, obtains second prediction under N kind topography base of each pixel
Label: C1,C2,...,CN;Step B3: by the second prediction label C of each pixel1,C2,...,CNWith coefficient of colligation ω1,
ω2,...,ωNWeighted sumObtain the reconstruct label of each pixel, thus obtain vision reconstruct image.
In sum, owing to have employed technique scheme, the invention has the beneficial effects as follows: before classification processes, preferentially
Carry out the screening of characteristic vector, reduce characteristic vector latitude, reduce computation complexity and the noise of reconstruct image of reconstruct;
Meanwhile, local loop can promote the degree of accuracy of visual pattern reconstruct further in based on the present invention 9.
Accompanying drawing explanation
Fig. 1 vision reconfiguration principle figure.
Fig. 2 training image sequence chart.
Fig. 3 tests image sequence figure.
What Fig. 4 visual pattern reconstructed is embodied as flow process.
Fig. 5 topography base transition diagram.
Fig. 6 visual pattern reconstruction result comparison diagram.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with embodiment and accompanying drawing, to this
Bright it is described in further detail.
Embodiment
Collection training sample and test sample:
See Fig. 2, allow subjects at first 28 seconds eye gaze tranquillization images (being easy to obtain the original position of fMRI signal);
Watch attentively the most successively 6 seconds be made up of random image stimulation figure (will random image with flicker checkerboard pattern (letter or geometry
Image is that black and white flashes checkerboard pattern, and background is Lycoperdon polymorphum Vitt) present to subjects) and 6 seconds tranquillization images, repeat 22 times;Finally
12 seconds eye gaze tranquillization images (being easy to obtain the final position of fMRI signal).Said process is performed 20 times, then can get
The fMRI signals of 440 times and corresponding random image (are stimulated the contrast of figure by the fMRI signal under 440 stimulating images
Figure) as training sample A.
See Fig. 3, allow first 28 seconds eye gaze tranquillization images of being tried;Watch test image (flicker in 12 seconds the most successively attentively
The stimulating image of checkerboard pattern) and 12 seconds tranquillization images, it is repeated 10 times;Latter 12 seconds eye gaze tranquillization images.By above-mentioned mistake
Cheng Zhihang 8 times, then can get the fMRI signal under 80 test images, using the fMRI signals of 80 times as test sample B.
Performing training and reconstruction processing respectively based on the training sample A obtained and test sample B, see Fig. 4, it specifically walks
Suddenly it is:
First, the stimulating image in training sample is carried out the conversion process of different scale, putting down of Ji Qu topography base
All contrast (for number and the ratio of chessboard sum of flicker chessboard defined in the window of topography's base place) is as current picture
The label of vegetarian refreshments, obtains the changing image corresponding with each topography base.See Fig. 5, use 9 kinds of Local maps as shown in Fig. 5-a
Carry out conversion process respectively as base, the window of i.e. 1 × 1 yardstick includes base φ1: the most only include current pixel point;1 × 2 yardstick
Window includes two base (φ2、φ3): i.e. include current pixel point and abutment points around;The window of 2 × 1 yardsticks includes two
Base (φ4、φ5): i.e. include current pixel point abutment points upper and lower with it;And 2 × 2 the window of yardstick include four base (φ6、
φ7、φ8、φ9): i.e. include that in the window of 2 × 2 of current pixel point, current pixel point lays respectively at the window upper left corner, upper right
Angle, the lower left corner, the lower right corner.Fig. 5-a gives the conversion schematic diagram under different base.I.e. use above-mentioned nine kinds of topography base φ1
~φ9Same stimulating image is carried out conversion process, respectively obtains nine width Transformation Graphs of correspondence.
Secondly, to training sample A, based on naive Bayes classifier and ten folding cross validation methods, visual pattern is built
Coefficient of colligation (the ω of each pixel1,ω2,...,ω9):
Under different topography's bases, use ten folding cross validation methods, respectively training sample is divided into 10 parts, 1
Part is as test data t1, and 9 parts, as training data s1, based on training data t1, use naive Bayes classifier to build pass
In first grader (characteristic vector is corresponding FMRI signal) of pixel position, then the FMRI signal of test data t1 is made
It is the input of the first grader, carries out Classification and Identification, obtain each pixel position the first pre-mark under N kind topography base
Sign, i.e. obtain many groups under s1, t1 (number depends on the test sample number included by t1) training data prediction labelsAfter ten folding cross processings, i.e. can get MA(sample number of training sample A) group
True pixel values (also can the claim true tag) C of the random image according to training sample AtrMark pre-with training data
SignSet up multiple linear regression model:Wherein ε represents
Residual error.Based on MAGroupUse the method for least square multiple linear equation matching to above-mentioned regression model,
Littleization mean square sesidualBy corresponding ω1~ωNAs the coefficient of colligation of each pixel of visual pattern, i.e.Wherein εjIn corresponding regression model
ε,C in corresponding regression modeltr, subscript j is sample identification symbol.
Coefficient of colligation (the ω of each pixel of visual pattern is then can get based on above-mentioned multiple linear regression model1,
ω2,...,ω9)。
Thirdly, classifying training sample based on pixel label, one is divided into five classes, and all kinds of label values is respectively
For: 0,0.25,0.5,0.75,1.Then use multi-class F-score feature selection approach that the fMRI signal of training sample A is entered
Row Feature Selection processes, i.e. basisCalculate the F value of each characteristic vector of fMRI signal respectively, then take front 5
Big F value is as Feature Selection result.Wherein SfarReflect different classes of between distance.ScloseReflection identical category between away from
From.SfarAnd ScloseComputing formula is respectivelyIts
Middle M is class number,Represent the average of all samples of characteristic vector;Represent in characteristic vector, belong to classification ciSon
The average of sample;niRepresent in characteristic vector, belong to classification ciThe length of subsample;Represent in characteristic vector, belong to class
Other ciSubsample in jth value.
Then, under different topography's bases, grader instruction is carried out based on the FMRI signal after changing image and screening
Practice, obtain the pixel the second grader at different topographies base, for the Tag Estimation of test sample.The present embodiment is adopted
Grader is built, to topography base φ with naive Bayes classifier1, training sample is divided into two classes, the picture of all kinds of correspondences
Vegetarian refreshments label is 0,1;To topography base φ2、φ3、φ4、φ5, respectively training sample is divided three classes, the picture of all kinds of correspondences
Vegetarian refreshments label is 0,0.5,1, to topography base φ6、φ7、φ8、φ9, respectively training sample is divided into five classes, all kinds of correspondences
Pixel label be 0,0.25,0.5,0.75,1.It is characterized vector with FMRI signal, builds the different local about pixel
Second grader of image base.Naive Bayes Classifier is a simple and efficient sorting technique.Naive Bayes Classification
Can predict that given sample belongs to the probability of a given class, calculate sample the most respectively and belong to each class probability, take general
Classification belonging to rate maximum.When setting up grader, the present invention can also use other machines study classification method to substitute
Naive Bayes Classification method, such as support vector machine (SVM), random forest (RF), sparse polynomial logistic regression (SMLR) etc.
Deng.For running the factor of time, these methods run the time much larger than the Nae Bayesianmethod operation time.In view of people
Class brain processes the such characteristic of visual information as quick as thought, uses Naive Bayes Classification method in the present embodiment.
Then, the FMRI signal of input training sample B, based on the K category feature screened, enters the FMRI signal of input
After row Feature Selection, as the input of Equations of The Second Kind grader, carry out classification differentiation, obtain each pixel position in difference local
The second prediction label under image basePre-as current pixel point training sample under each topography base
Mark label.
Finally, the training sample of same pixel position is predicted labelWith reconstruction parameter ω1,
ω2,...,ω9Being weighted sues for peace obtains the reconstruct label of current pixel point position, i.e. topography's reconstruct
Thus obtain the heavy pixel value of each pixel in visual pattern, i.e. obtain the visual pattern of reconstruct.
Fig. 6 gives the Contrast on effect of the present invention and the reconstruct visual pattern of existing mode, and wherein Fig. 6-a is for using SMLR
The reconstruct visual pattern of mode, 6-a is the reconstruct visual pattern of the present invention.From figure, the result noise that the present invention obtains shows
Write and reduce.
The above, the only detailed description of the invention of the present invention, any feature disclosed in this specification, unless especially
Narration, all can be by other equivalences or have the alternative features of similar purpose and replaced;Disclosed all features or all sides
Method or during step, in addition to mutually exclusive feature and/or step, all can be combined in any way.
Claims (6)
1. a high-efficient noise-reducing visual pattern reconstructing method, it is characterised in that comprise the following steps:
A. training step:
Step A1: input training sample, described training sample includes the stimulating image FMRI signal in brain visual area, Yi Jiyu
The contrast figure of FMRI signal stimulating image one to one, wherein stimulating image is flicker checkerboard pattern;
Step A2: use N kind topography base that the contrast figure of stimulating image is carried out conversion process, obtain same contrast figure
N width changing image, wherein N be more than 1:
Based on current topography base, respectively each pixel of stimulating image is carried out conversion process, take the flat of topography's base
All contrast is as the label of current pixel point, obtains the changing image of current topography base, and wherein topography's base is flat
All contrasts are ratio total with chessboard for the number of flicker chessboard defined in the window of place;
Step A3: the coefficient of colligation of the structure each pixel of visual pattern:
Step A3-1: each pixel position to the contrast figure of stimulating image, the training of acquisition N kind topography base is pre-respectively
Mark label:
Training sample is divided into two subsets, and a subset is as training data, and a subset is as test data, different
Under topography's base, based on changing image and FMRI signal, training data is carried out classifier training, obtain each pixel not
With the first grader of topography's base, wherein the classification of the first grader is the taking of average contrast of each topography base
Value;
By the first grader, test data are carried out Classification and Identification, obtain each pixel position under N kind topography base
First prediction labelWherein i=1,2 ..., N;
Step A3-2: according to formulaIn multiple test data, the numerical value minimum of residual epsilon will be made
ω1,ω2,...,ωNAs the coefficient of colligation of current pixel point position, thus obtain combining of each pixel of visual pattern and be
Number, wherein CtrRepresent the pixel value of the contrast figure of stimulating image;
Step A4: the value of average contrast based on N kind topography base divides classification, training sample is carried out classification and draws
Point;And the characteristic vector of the FMRI signal of training sample is carried out Feature Selection, take the feature that front K classification judgement ability is maximum
Vector is as Feature Selection result, and wherein K is preset value;
Under different topography's bases, carry out classifier training based on the FMRI signal after changing image and screening, obtain picture
Vegetarian refreshments is at the second grader of different topographies base, and wherein the classification of the second grader is the most right of N kind topography base
Value than degree;
Step B: image reconstruction
Step B1: input FMRI signal to be reconstructed, and the selection result based on step A4 treats the spy of FMRI signal of reconstruct
Levying and screen, K characteristic vector after screening is as the input of the second grader;
Step B2: recognition result based on the second grader, obtains each pixel second pre-mark under N kind topography base
Sign: C1,C2,...,CN;Step B3: by the second prediction label C of each pixel1,C2,...,CNWith coefficient of colligation ω1,
ω2,...,ωNWeighted sum obtain the reconstruct label of each pixel, thus obtain vision reconstruct image.
2. the method for claim 1, it is characterised in that topography's base includes nine kinds, respectively:
Topography base φ1: only include current pixel point 1 × 1 window;
Topography base φ2: include current pixel point abutment points right with it 1 × 2 window;
Topography base φ3: include current pixel point abutment points left with it 1 × 2 window;
Topography base φ4: include current pixel point with its on abutment points 2 × 1 window;;
Topography base φ5: include that current pixel point and its descend the window of the 2 × 1 of abutment points;
Topography base φ6: include current pixel point 2 × 2 window, wherein current pixel point is in the upper left corner of window;
Topography base φ7: include current pixel point 2 × 2 window, wherein current pixel point is in the upper right corner of window;
Topography base φ8: include current pixel point 2 × 2 window, wherein current pixel point is in the lower left corner of window;
Topography base φ9: include current pixel point 2 × 2 window, wherein current pixel point is in the lower right corner of window;
In step A4, training sample is divided into five classes, the contrast of all kinds of correspondences are respectively as follows: 0,0.25,0.5,0.75,1;
The classification information of different topographies base is:
Topography base φ1Including two classes, all kinds of contrasts is respectively as follows: 0,1;
Topography base φ2、φ3、φ4、φ5Including three classes, all kinds of contrasts is respectively as follows: 0,0.5,1;
Topography base φ6、φ7、φ8、φ9Including five classes, all kinds of contrasts is respectively as follows: 0,0.25,0.5,0.75,1.
3. method as claimed in claim 1 or 2, it is characterised in that in described step A4, use multi-class F-score feature
Selection mode carries out Feature Selection.
4. method as claimed in claim 1 or 2, it is characterised in that use naive Bayes classifier to obtain first, second point
Class device.
5. method as claimed in claim 1 or 2, it is characterised in that in described step A3-1, uses ten folding interior extrapolation methods to obtain the
One prediction label
6. method as claimed in claim 1 or 2, it is characterised in that in step A3-2, uses method of least square to obtain associating ginseng
Number ω1,ω2,...,ωN。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610589708.4A CN106251299B (en) | 2016-07-25 | 2016-07-25 | A kind of high-efficient noise-reducing visual pattern reconstructing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610589708.4A CN106251299B (en) | 2016-07-25 | 2016-07-25 | A kind of high-efficient noise-reducing visual pattern reconstructing method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106251299A true CN106251299A (en) | 2016-12-21 |
CN106251299B CN106251299B (en) | 2019-05-10 |
Family
ID=57603922
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610589708.4A Active CN106251299B (en) | 2016-07-25 | 2016-07-25 | A kind of high-efficient noise-reducing visual pattern reconstructing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106251299B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108573512A (en) * | 2018-03-21 | 2018-09-25 | 电子科技大学 | A kind of complicated visual pattern reconstructing method based on depth encoding and decoding veneziano model |
CN108960073A (en) * | 2018-06-05 | 2018-12-07 | 大连理工大学 | Cross-module state image steganalysis method towards Biomedical literature |
CN113362408A (en) * | 2021-05-11 | 2021-09-07 | 山东师范大学 | Bayes reconstruction method and system for brain activity multi-scale local contrast image |
CN114469009A (en) * | 2022-03-18 | 2022-05-13 | 电子科技大学 | Facial pain expression grading evaluation method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1777818A (en) * | 2003-04-24 | 2006-05-24 | 皇家飞利浦电子股份有限公司 | Fibre tracking magnetic resonance imaging |
CN101224114A (en) * | 2008-01-25 | 2008-07-23 | 西安交通大学 | High dynamic range regenerating method of X-ray image based on scale space decomposition |
CN103247028A (en) * | 2013-03-19 | 2013-08-14 | 广东技术师范学院 | Multi-hypothesis prediction block compressed sensing image processing method |
US20140023253A1 (en) * | 2012-07-20 | 2014-01-23 | Jan S. Hesthaven | Image Reconstruction from Incomplete Fourier Measurements and Prior Edge Information |
US20140180060A1 (en) * | 2012-12-17 | 2014-06-26 | Todd Parrish | Methods and Systems for Automated Functional MRI in Clinical Applications |
CN105068644A (en) * | 2015-07-24 | 2015-11-18 | 山东大学 | Method for detecting P300 electroencephalogram based on convolutional neural network |
-
2016
- 2016-07-25 CN CN201610589708.4A patent/CN106251299B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1777818A (en) * | 2003-04-24 | 2006-05-24 | 皇家飞利浦电子股份有限公司 | Fibre tracking magnetic resonance imaging |
CN101224114A (en) * | 2008-01-25 | 2008-07-23 | 西安交通大学 | High dynamic range regenerating method of X-ray image based on scale space decomposition |
US20140023253A1 (en) * | 2012-07-20 | 2014-01-23 | Jan S. Hesthaven | Image Reconstruction from Incomplete Fourier Measurements and Prior Edge Information |
US20140180060A1 (en) * | 2012-12-17 | 2014-06-26 | Todd Parrish | Methods and Systems for Automated Functional MRI in Clinical Applications |
CN103247028A (en) * | 2013-03-19 | 2013-08-14 | 广东技术师范学院 | Multi-hypothesis prediction block compressed sensing image processing method |
CN105068644A (en) * | 2015-07-24 | 2015-11-18 | 山东大学 | Method for detecting P300 electroencephalogram based on convolutional neural network |
Non-Patent Citations (4)
Title |
---|
SUTAO SONG 等: "Bayesian reconstruction of multiscale local contrast images from brain activity brain activity", 《JOURNAL OF NEUROSCIENCE METHODS 》 * |
YOICHI MIYAWAKI 等: "Visual Image Reconstruction from Human Brain Activity using a Combination of Multiscale Local Image Decoders", 《NEURON》 * |
宋素涛 等: "基于fMRI的视觉信息解码研究进展", 《济南大学学报(自然科学版)》 * |
谢娟英 等: "基于改进的F-score与支持向量机的特征选择方法", 《计算机应用》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108573512A (en) * | 2018-03-21 | 2018-09-25 | 电子科技大学 | A kind of complicated visual pattern reconstructing method based on depth encoding and decoding veneziano model |
CN108573512B (en) * | 2018-03-21 | 2021-04-30 | 电子科技大学 | Complex visual image reconstruction method based on depth coding and decoding dual model |
CN108960073A (en) * | 2018-06-05 | 2018-12-07 | 大连理工大学 | Cross-module state image steganalysis method towards Biomedical literature |
CN108960073B (en) * | 2018-06-05 | 2020-07-24 | 大连理工大学 | Cross-modal image mode identification method for biomedical literature |
CN113362408A (en) * | 2021-05-11 | 2021-09-07 | 山东师范大学 | Bayes reconstruction method and system for brain activity multi-scale local contrast image |
CN114469009A (en) * | 2022-03-18 | 2022-05-13 | 电子科技大学 | Facial pain expression grading evaluation method |
Also Published As
Publication number | Publication date |
---|---|
CN106251299B (en) | 2019-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Noreen et al. | A deep learning model based on concatenation approach for the diagnosis of brain tumor | |
Farooq et al. | A deep CNN based multi-class classification of Alzheimer's disease using MRI | |
Işın et al. | Review of MRI-based brain tumor image segmentation using deep learning methods | |
Cheng et al. | CNNs based multi-modality classification for AD diagnosis | |
Jabason et al. | Classification of Alzheimer’s disease from MRI data using an ensemble of hybrid deep convolutional neural networks | |
Yu et al. | Hybrid dermoscopy image classification framework based on deep convolutional neural network and Fisher vector | |
CN106251299A (en) | A kind of high-efficient noise-reducing visual pattern reconstructing method | |
CN106408001A (en) | Rapid area-of-interest detection method based on depth kernelized hashing | |
CN106096636A (en) | A kind of Advancement Type mild cognition impairment recognition methods based on neuroimaging | |
CN106127263A (en) | The human brain magnetic resonance image (MRI) classifying identification method extracted based on three-dimensional feature and system | |
CN111714118A (en) | Brain cognition model fusion method based on ensemble learning | |
CN109816630A (en) | FMRI visual coding model building method based on transfer learning | |
Ortiz et al. | Learning longitudinal MRI patterns by SICE and deep learning: Assessing the Alzheimer’s disease progression | |
CN109902682A (en) | A kind of mammary gland x line image detection method based on residual error convolutional neural networks | |
Ben-Cohen et al. | Anatomical data augmentation for CNN based pixel-wise classification | |
Aloyayri et al. | Breast cancer classification from histopathological images using transfer learning and deep neural networks | |
Divya et al. | A deep transfer learning framework for multi class brain tumor classification using MRI | |
Liu et al. | Multi-LSTM networks for accurate classification of attention deficit hyperactivity disorder from resting-state fMRI data | |
Qian et al. | 3D automatic segmentation of brain tumor based on deep neural network and multimodal MRI images | |
Sharma et al. | A Suitable Approach for Classifying Skin Disease Using Deep Convolutional Neural Network | |
Pusparani et al. | Diagnosis of Alzheimer’s disease using convolutional neural network with select slices by landmark on Hippocampus in MRI images | |
Castro et al. | Generation of synthetic structural magnetic resonance images for deep learning pre-training | |
Rezaei et al. | Brain abnormality detection by deep convolutional neural network | |
Yeung et al. | Pipeline comparisons of convolutional neural networks for structural connectomes: predicting sex across 3,152 participants | |
CN109146005A (en) | A kind of brain cognitive ability analysis method based on rarefaction representation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |