CN103310227A - Automatic window width and window level extraction method based on neural network - Google Patents
Automatic window width and window level extraction method based on neural network Download PDFInfo
- Publication number
- CN103310227A CN103310227A CN2012100716830A CN201210071683A CN103310227A CN 103310227 A CN103310227 A CN 103310227A CN 2012100716830 A CN2012100716830 A CN 2012100716830A CN 201210071683 A CN201210071683 A CN 201210071683A CN 103310227 A CN103310227 A CN 103310227A
- Authority
- CN
- China
- Prior art keywords
- image
- window
- class
- window width
- neural network
- 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.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an automatic window width and window level extraction method based on neural network. The automatic window width and window level extraction method classifies MR (magnetic resonance) images by utilizing an adaptive K clustering method, and comprises the following online training steps of (e) extracting histogram characters and space information characters of new MR images; (f) according to the histogram characters and the space information characters of the new MR images, classifying the new MR images by utilizing the adaptive K clustering method; and (g) comparing each class of the classified new MR images with each class of the current trained images, and firstly calculating the similarity of each class of the images with each class of the current trained images; if no similarity exits, adding a new class based on the original classes; and if the similarity exists, judging whether the window width and the window level of the class of the images are same as the golden standards of the window width and the window level of the class of the current trained images or not; if the window width and the window level of the class of the images are same as the golden standards of the window width and the window level of the class of the current trained images, not adding the new class; and if the window width and the window level of the class of the images are not same as the golden standards of the window width and the window level of the class of the current trained images, adding a new class based on the original classes. According to the automatic window width and window level extraction method based on neural network, which is disclosed by the invention, the online training can be automatically realized.
Description
Technical field
The present invention relates to a kind of Medical Image Processing technology, relate in particular to a kind of automatic window width and window level extracting method based on neural network.
Background technology
At present, automatic window width and window level extracting method epineural network use also fewer.Paper 1 (Ohliashi A, Yamada S, Haruki K, Hatano H, Fujii Y, Yamaguchi K, Ogata H.Automatic adjustment of display window for MR images using a neural network, Proceeding of SPIE, vol.1444, Image Capture, Formatting and Display, 1991.p.63-74) only used a kind of neural network, it can not well cover the MR image of all kinds.Paper 2 (Lai SH, Fang M.A hierarchical neural network algorithm for robust and automatic windowing of MR images.Artif Intell Med 2000; 19 (2): 97-119) used two-layer neural network, as shown in Figure 1, comprised the steps: step S101, loaded the MR image; Step S102, histogram feature and the spatial information of extraction all images; Step S103 treats dissimilar training images with self-organizing map neural network (SOM) and carries out rough sort, and this step can be reduced the complicacy of input data; Step S104 is with corresponding radial basis function neural network (RBF) and the bimodulus linear neural network (BLE) of goldstandard window width and window level training of each class image correspondence.Though it has well solved the problem of paper 1, when a kind of new MR sequence image occurring, perhaps to same kind MR image, when the window width and window level of physician preference changed, the result who obtains was not ideal.Because the cluster centre that the SOM clustering method obtains is the optimum mapping of all samples, so only providing under the prerequisite of original training sample, could train together with new training sample and to obtain new cluster centre, improve as need, all need to train again for the corresponding neural network of all cluster centres that newly obtain, thereby can't realize online automatic training, need whole neural network is trained again.Paper 3 (Lai SH, Fang M.An adaptive window width/center adjustment system with online training capabilities for MR images.Artif Intell Med 2005; 33 (2): 89-101) be online training, it changes the direction of all window width and window levels to physician preference before training, trains whole neural network then again.Need artificial intervention in the process of online training, the problem that it and paper 2 exist same needs to train again.
This shows that the automatic extractive technique of existing window width and window level based on neural network is running into new MR image sequence, or the window width and window level of certain parts of images sequence need train whole neural network when changing again, use inconvenience and influence efficient.
Summary of the invention
Technical matters to be solved by this invention provides a kind of automatic window width and window level extracting method based on neural network, when the window width and window level that new MR image sequence or a certain kind occur changes, only training increases the neural network of types of image newly, other neural networks remain unchanged, thereby online training can be realized automatically.
The present invention solves the problems of the technologies described above the technical scheme that adopts to provide a kind of automatic window width and window level extracting method based on neural network, comprises the steps: a) to load the MR image; B) histogram feature and the spatial information feature of extraction MR image; C) according to histogram feature and the spatial information feature of MR image, utilize self-adaptation K clustering method that all MR images are classified; D) with the window width and window level information of such image feature information and image such radial basis function neural network is trained respectively; E) change when producing new MR image when the window width and window level that loads new MR image or certain class image, extract histogram feature and the spatial information feature of new MR image; F) according to histogram feature and the spatial information feature of new MR image, utilize described self-adaptation K clustering method that new MR image is classified; G) the new MR image of sorted every class is compared with existing every class image of having trained, calculate every class image and existing every class image similarity of having trained earlier, if it is dissimilar, then increase a new class at former basis of classification, if similar, judge again whether the window width and window level of such image is the same with the window width and window level goldstandard of existing training image, if it is the same, do not increase new class, if different, then increase a new class at former basis of classification; H) loading the MR image tests; I) extract histogram feature and the spatial information feature of testing the MR image; J) according to histogram feature and the spatial information feature of test MR image, calculate and the histogram feature of each cluster centre and the similarity of spatial information feature, obtain a plurality of cluster centres the most similar to it; K) utilize the radial base neural net of described a plurality of cluster centre correspondences to calculate a plurality of window width and window levels respectively; L) merge described a plurality of window width and window levels, export final window width and window position.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, described histogram feature extracts after utilizing small echo to change dimensionality reduction.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, the extracting method of described histogram feature is as follows: 1) according to its histogram of MR image calculation; 2) histogram is done the sampling that descends and handle that to obtain a size be 64 vector; 3) described vector being done twice wavelet transformation, to obtain a size be 16 histogram feature vector.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, described spatial information feature comprises average and the variance of the gray-scale value of different spatial in the MR image.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, described spatial information Feature Extraction method is as follows: 1) image evenly is divided into the 4*4 zone, calculates average and the variance of all pixel values respectively in each described zone; 2) calculating a size according to described average and variance is 16 space characteristics vector.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, described self-adaptation K clustering method is the MR image to be classified behind 32 the vector according to merging histogram feature and spatial information feature and being normalized into size.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, described similarity is judged every class image and the similarity that has every class image of having trained now by the Euclidean distance between 32 features calculating two class picture centre correspondences.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, when the described Euclidean distance of every class image greater than 0.5 the time, be judged as dissmilarity.
The above-mentioned automatic window width and window level extracting method based on neural network, wherein, the window width and window level of described new class image is compared with the existing window width and window level of training image, and window width or window position change above 15%, are judged as dissmilarity.
The present invention contrasts prior art following beneficial effect: the automatic window width and window level extracting method based on neural network provided by the invention, use self-adaptation K clustering method that the MR image is classified, and with the window width and window level information of such image feature information and image such radial basis function neural network is trained respectively, when the window width and window level that new MR image sequence or a certain kind occur changes, self-adaptation K clustering method can add new cluster centre under the situation that does not change original cluster centre, thereby only training increases the corresponding neural network of types of image (newly-increased cluster centre) newly, other neural networks remain unchanged, thereby online training can be realized automatically, use on-line study need not to provide under the situation of original training sample, only improve existing model with new (use habit of being close to the users) data.By the new data of continuous increase, thereby make the window width and window level of automatic calculating meet user's requirement more.
Description of drawings
Fig. 1 is based on the process flow diagram of training stage in the automatic window width and window level extracting method of neural network in the prior art;
Fig. 2 is the process flow diagram that the present invention is based on the training stage in the automatic window width and window level extracting method of neural network;
Fig. 3 is the process flow diagram that the present invention is based on the online training stage in the automatic window width and window level extracting method of neural network;
Fig. 4 is the process flow diagram that the present invention is based on operational phase in the automatic window width and window level extracting method of neural network.
Embodiment
The invention will be further described below in conjunction with drawings and Examples.
Fig. 2 is the process flow diagram based on the training stage in the automatic window width and window level extracting method of neural network of the present invention.
See also Fig. 2, the automatic window width and window level extracting method based on neural network provided by the invention during for the original sample training, comprises the steps:
Step S201 loads the MR image.
Step S202, histogram feature and the spatial information feature of extraction MR image; Described histogram feature comprises the distribution of gray-scale value in the MR image, utilizes the wavelet transformation dimensionality reduction before the extraction, and the histogram feature extracting method is as follows: 1) according to its histogram of MR image calculation; 2) histogram is done the sampling that descends and handle that to obtain a size be 64 vector; 3) described vector being done twice wavelet transformation, to obtain a size be 16 histogram feature vector.Described spatial information feature comprises average and the variance of the gray-scale value of different spatial in the MR image; Computing method are as follows: 1) image evenly is divided into the individual zone of 16 (4*4), calculates average and the variance of all pixel values in this zone respectively in each described zone; 2) calculating a size according to described average and variance is 16 space characteristics vector.
Step S203, histogram feature and spatial information feature according to the MR image, namely according to merge histogram feature and spatial information feature and be normalized into size be a vector of 32 after, utilize self-adaptation K cluster (Adaptive K-Means) method that all MR images are classified.
Step S204 to sorted every class image among the step S203, uses the window width and window level information of such image feature information and image that such radial basis function neural network (RBF) is trained respectively.
Fig. 3 is the process flow diagram based on online training in the automatic window width and window level extracting method of neural network of the embodiment of the invention.
See also Fig. 3, the automatic window width and window level extracting method based on neural network provided by the invention when the window width and window level that loads new MR image or certain class image changes when producing new MR image, comprises following online training step:
Step S301 loads new MR image;
Step S302 extracts histogram feature and the spatial information feature of new MR image; Described histogram feature comprises the distribution of gray-scale value in the MR image, utilizes the wavelet transformation dimensionality reduction before the extraction, and the histogram feature extracting method is as follows: 1) according to its histogram of MR image calculation; 2) histogram is done the sampling that descends and handle that to obtain a size be 64 vector; 3) described vector being done twice wavelet transformation, to obtain a size be 16 histogram feature vector.Described spatial information feature comprises average and the variance of the gray-scale value of different spatial in the MR image; Computing method are as follows: 1) image evenly is divided into the individual zone of 16 (4*4), calculates average and the variance of all pixel values in this zone respectively in each described zone; 2) calculating a size according to described average and variance is 16 space characteristics vector.
Step S303, according to histogram feature and the spatial information feature of new MR image, namely according to merge histogram feature and spatial information feature and be normalized into size be a vector of 32 after, utilize self-adaptation K clustering method that new MR image is classified.
Step S304, the MR image new to the sorted every class of step S303 compares with existing every class image of having trained, calculate every class image and existing every class image similarity of having trained earlier, if dissimilar, then increase a new class at former basis of classification, if it is similar, whether the window width and window level of judging such image again is the same with the window width and window level goldstandard of existing training image, if the same, do not increase new class, if different, then increase a new class at former basis of classification.
Among the above-mentioned steps S304, judge every class image and the similarity that has every class image of having trained now by the Euclidean distance between 32 features calculating two class picture centre correspondences.When the Euclidean distance between the feature of every class image and the existing corresponding cluster centre of every class image of having trained greater than 0.5 the time, be judged as dissmilarity, increase new class.If image similarity is compared with the existing window width and window level of training image when the window width and window level of new class image, window width or window position changed above 15% o'clock, were judged as dissmilarity, increased new class.
Fig. 4 is the process flow diagram based on operational phase in the automatic window width and window level extracting method of neural network of the embodiment of the invention.
See also Fig. 4, the automatic window width and window level extracting method based on neural network provided by the invention when loading test MR image, comprises following calculation procedure:
Step S401 loads test MR image;
Step S402 extracts histogram feature and the spatial information feature of testing the MR image;
Step S403 according to histogram feature and the spatial information feature of test MR image, calculates and the histogram feature of each cluster centre and the similarity of spatial information feature, obtains a plurality of cluster centres the most similar to it; The number preferable range of cluster centre is 2-8, and the best is 5.
Step S404 utilizes the radial base neural net of described a plurality of cluster centre correspondences to calculate a plurality of window width and window levels respectively.
Step S405 merges described a plurality of window width and window level, exports final window width and window position.
The extracting method of window width and window level in sum, when the window width and window level that new MR image sequence or a certain kind occur changes, self-adaptation K clustering method can add new cluster centre under the situation that does not change original cluster centre, thereby only training increases the corresponding neural network of types of image (newly-increased cluster centre) newly, other neural networks remain unchanged, thereby online training can be realized automatically, use on-line study need not to provide under the situation of original training sample, only improve existing model with new (use habit of being close to the users) data.
In addition, the present invention has realized than the littler average error (Mean of the window width and window level extracting method of prior art
WW, Mean
WL), average error is defined as follows:
Wherein:
Error
WW(i)＝|WW(i)-WW
gt(i)|，i∈{1，2，...，N}
Error
WL(i)＝|WL(i)-WL
gt(i)|，i∈{1，2，...，N}
N is the positive integer greater than 1.
With 2 disclosed methods of paper in background technology object as a comparison, the test result of 5863 width of cloth training images is seen Table 1,200 width of cloth newly add image measurement and the results are shown in Table 2.
The test result of table 15863 width of cloth training image
| Mean WL(％) | Mean WW(％) | |
| Existing method | 14.33 | 14.92 |
| The inventive method | 14.21 | 14.01 |
Table 2200 width of cloth newly adds the image measurement result
| Mean WL(％) | Mean WW(％) | |
| Existing method | 18.09 | 17.86 |
| The inventive method | 14.57 | 13.88 |
As can be seen from the above table, regardless of being to training image or newly adding image, use the comparison average error (Mean of the inventive method
WW, Mean
WL) all to be lower than existing method.
Though the present invention discloses as above with preferred embodiment; right its is not in order to limiting the present invention, any those skilled in the art, without departing from the spirit and scope of the present invention; when can doing a little modification and perfect, so protection scope of the present invention is when with being as the criterion that claims were defined.
Claims (9)
1. the automatic window width and window level extracting method based on neural network is characterized in that, comprises the steps:
A) load the MR image;
B) histogram feature and the spatial information feature of extraction MR image;
C) according to histogram feature and the spatial information feature of MR image, utilize self-adaptation K clustering method that all MR images are classified;
D) with the window width and window level information of such image feature information and image such radial basis function neural network is trained respectively;
E) change when producing new MR image when the window width and window level that loads new MR image or certain class image, extract histogram feature and the spatial information feature of new MR image;
F) according to histogram feature and the spatial information feature of new MR image, utilize described self-adaptation K clustering method that new MR image is classified;
G) the new MR image of sorted every class is compared with existing every class image of having trained, calculate every class image and existing every class image similarity of having trained earlier, if it is dissimilar, then increase a new class at former basis of classification, if similar, judge again whether the window width and window level of such image is the same with the window width and window level goldstandard of existing training image, if it is the same, do not increase new class, if different, then increase a new class at former basis of classification;
H) loading the MR image tests;
I) extract histogram feature and the spatial information feature of testing the MR image;
J) according to histogram feature and the spatial information feature of test MR image, calculate and the histogram feature of each cluster centre and the similarity of spatial information feature, obtain a plurality of cluster centres the most similar to it;
K) utilize the radial base neural net of described a plurality of cluster centre correspondences to calculate a plurality of window width and window levels respectively;
L) merge described a plurality of window width and window levels, export final window width and window position.
2. the automatic window width and window level extracting method based on neural network as claimed in claim 1 is characterized in that, described histogram feature extracts after utilizing small echo to change dimensionality reduction.
3. the automatic window width and window level extracting method based on neural network as claimed in claim 2 is characterized in that the extracting method of described histogram feature is as follows: 1) according to its histogram of MR image calculation; 2) histogram is done the sampling that descends and handle that to obtain a size be 64 vector; 2) described vector being done twice wavelet transformation, to obtain a size be 16 histogram feature vector.
4. the automatic window width and window level extracting method based on neural network as claimed in claim 1 is characterized in that described spatial information feature comprises average and the variance of the gray-scale value of different spatial in the MR image.
5. the automatic window width and window level extracting method based on neural network as claimed in claim 4, it is characterized in that, described spatial information Feature Extraction method is as follows: 1) image evenly is divided into the 4*4 zone, calculates average and the variance of all pixel values respectively in each described zone; 2) calculating a size according to described average and variance is 16 space characteristics vector.
6. the automatic window width and window level extracting method based on neural network as claimed in claim 1, it is characterized in that described self-adaptation K clustering method is the MR image to be classified behind 32 the vector according to merging histogram feature and spatial information feature and being normalized into size.
7. the automatic window width and window level extracting method based on neural network as claimed in claim 1 is characterized in that described similarity judges the similarity of every class image and existing every class image of having trained by the Euclidean distance between 32 features calculating two class picture centre correspondences.
8. the automatic window width and window level extracting method based on neural network as claimed in claim 7 is characterized in that, when the described Euclidean distance of every class image greater than 0.5 the time, be judged as dissmilarity.
9. the automatic window width and window level extracting method based on neural network as claimed in claim 1 is characterized in that, the window width and window level of described new class image is compared with the existing window width and window level of training image, and window width or window position change above 15%, are judged as dissmilarity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012100716830A CN103310227A (en) | 2012-03-16 | 2012-03-16 | Automatic window width and window level extraction method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012100716830A CN103310227A (en) | 2012-03-16 | 2012-03-16 | Automatic window width and window level extraction method based on neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103310227A true CN103310227A (en) | 2013-09-18 |
Family
ID=49135421
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012100716830A Pending CN103310227A (en) | 2012-03-16 | 2012-03-16 | Automatic window width and window level extraction method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103310227A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105938558A (en) * | 2015-03-06 | 2016-09-14 | 松下知识产权经营株式会社 | Learning method |
| CN108229691A (en) * | 2018-01-24 | 2018-06-29 | 北京墨丘科技有限公司 | Data validity determines method, apparatus, electronic equipment and storage medium |
| CN108334733A (en) * | 2017-09-01 | 2018-07-27 | 上海联影医疗科技有限公司 | Medical image display method, display system and computer readable storage medium |
| CN108537794A (en) * | 2018-04-19 | 2018-09-14 | 上海联影医疗科技有限公司 | Medical image processing method, device and computer readable storage medium |
| CN109035203A (en) * | 2018-06-25 | 2018-12-18 | 青岛海信医疗设备股份有限公司 | Medical image processing method, device, equipment and storage medium |
| CN109273074A (en) * | 2018-11-27 | 2019-01-25 | 数坤（北京）网络科技有限公司 | A kind of network model method of adjustment and equipment for medical image |
| CN110334731A (en) * | 2019-05-09 | 2019-10-15 | 云南大学 | A kind of the extraction of spatial information method, apparatus and electronic equipment of spectrum picture |
| CN112686874A (en) * | 2020-12-31 | 2021-04-20 | 苏州朗润医疗系统有限公司 | MRI (magnetic resonance imaging) self-adaptive window width position extraction method based on convolutional neural network |
| WO2021115356A1 (en) * | 2020-05-15 | 2021-06-17 | 平安科技（深圳）有限公司 | Adaptive window width and window center position adjustment method, apparatus, computer system, and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5995644A (en) * | 1997-06-30 | 1999-11-30 | Siemens Corporate Research, Inc. | Robust and automatic adjustment of display window width and center for MR images |
| CN102135606A (en) * | 2010-12-13 | 2011-07-27 | 电子科技大学 | KNN (K-Nearest Neighbor) sorting algorithm based method for correcting and segmenting grayscale nonuniformity of MR (Magnetic Resonance) image |
-
2012
- 2012-03-16 CN CN2012100716830A patent/CN103310227A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5995644A (en) * | 1997-06-30 | 1999-11-30 | Siemens Corporate Research, Inc. | Robust and automatic adjustment of display window width and center for MR images |
| CN102135606A (en) * | 2010-12-13 | 2011-07-27 | 电子科技大学 | KNN (K-Nearest Neighbor) sorting algorithm based method for correcting and segmenting grayscale nonuniformity of MR (Magnetic Resonance) image |
Non-Patent Citations (4)
| Title |
|---|
| AKINAMI OHHASHI ETC.: "Automatic adjustment of display window (gray level) for MR images using a neural network", 《PROCEEDING OF SPIE》 * |
| SHANG-HONG LAI ETC.: "An adaptive window width/center adjustment system with online training capabilities for MR images", 《ARTIFICIAL INTELLIGENCE IN MEDICINE 2005》 * |
| SHANG-HONG LAI: "A hierarchical neural network algorithm for robust and automatic windowing of MR images", 《ARTIFICIAL INTELLIGENCE IN MEDICINE 2000》 * |
| 吴功宜等: "《网络安全高级软件编程技术》", 30 April 2010, 清华大学出版社 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105938558B (en) * | 2015-03-06 | 2021-02-09 | 松下知识产权经营株式会社 | Learning method |
| CN105938558A (en) * | 2015-03-06 | 2016-09-14 | 松下知识产权经营株式会社 | Learning method |
| CN108334733A (en) * | 2017-09-01 | 2018-07-27 | 上海联影医疗科技有限公司 | Medical image display method, display system and computer readable storage medium |
| CN108334733B (en) * | 2017-09-01 | 2021-03-16 | 上海联影医疗科技股份有限公司 | Medical image display method, display system and computer-readable storage medium |
| CN108229691A (en) * | 2018-01-24 | 2018-06-29 | 北京墨丘科技有限公司 | Data validity determines method, apparatus, electronic equipment and storage medium |
| CN108537794A (en) * | 2018-04-19 | 2018-09-14 | 上海联影医疗科技有限公司 | Medical image processing method, device and computer readable storage medium |
| CN108537794B (en) * | 2018-04-19 | 2021-09-21 | 上海联影医疗科技股份有限公司 | Medical image data processing method, apparatus and computer readable storage medium |
| CN109035203A (en) * | 2018-06-25 | 2018-12-18 | 青岛海信医疗设备股份有限公司 | Medical image processing method, device, equipment and storage medium |
| CN109273074B (en) * | 2018-11-27 | 2019-12-10 | 数坤（北京）网络科技有限公司 | Network model adjusting method and equipment for medical image |
| CN109273074A (en) * | 2018-11-27 | 2019-01-25 | 数坤（北京）网络科技有限公司 | A kind of network model method of adjustment and equipment for medical image |
| CN110334731A (en) * | 2019-05-09 | 2019-10-15 | 云南大学 | A kind of the extraction of spatial information method, apparatus and electronic equipment of spectrum picture |
| CN110334731B (en) * | 2019-05-09 | 2022-04-12 | 云南大学 | Spectral image spatial information extraction method and device and electronic equipment |
| WO2021115356A1 (en) * | 2020-05-15 | 2021-06-17 | 平安科技（深圳）有限公司 | Adaptive window width and window center position adjustment method, apparatus, computer system, and storage medium |
| CN112686874A (en) * | 2020-12-31 | 2021-04-20 | 苏州朗润医疗系统有限公司 | MRI (magnetic resonance imaging) self-adaptive window width position extraction method based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103310227A (en) | Automatic window width and window level extraction method based on neural network | |
| CN109344736B (en) | Static image crowd counting method based on joint learning | |
| CN105095856B (en) | Face identification method is blocked based on mask | |
| CN103098090B (en) | Multiparameter three-dimensional magnetic resonance imaging brain tumor partition method | |
| CN106408030B (en) | SAR image classification method based on middle layer semantic attribute and convolutional neural networks | |
| CN111582008B (en) | Device and method for training classification model and device for classifying by using classification model | |
| CN106951825A (en) | A kind of quality of human face image assessment system and implementation method | |
| CN106846316A (en) | A kind of GIS inside typical defect automatic distinguishing method for image | |
| CN109615014A (en) | A kind of data sorting system and method based on the optimization of KL divergence | |
| CN103984959A (en) | Data-driven and task-driven image classification method | |
| CN104834940A (en) | Medical image inspection disease classification method based on support vector machine (SVM) | |
| US20180041765A1 (en) | Compact video representation for video event retrieval and recognition | |
| Percannella et al. | A classification-based approach to segment HEp-2 cells | |
| CN103136504A (en) | Face recognition method and device | |
| CN104680144A (en) | Lip language recognition method and device based on projection extreme learning machine | |
| CN102495901B (en) | Method for keeping balance of implementation class data through local mean | |
| CN104050361A (en) | Intelligent analysis early warning method for dangerousness tendency of prison persons serving sentences | |
| CN101251896B (en) | Object detecting system and method based on multiple classifiers | |
| CN107392221B (en) | Training method of classification model, and method and device for classifying OCR (optical character recognition) results | |
| CN109858570A (en) | Image classification method and system, computer equipment and medium | |
| CN108764361A (en) | The operating mode's switch method of beam type oil pumping machine indicating diagram based on integrated study | |
| CN108664969A (en) | Landmark identification method based on condition random field | |
| CN104217213A (en) | Medical image multi-stage classification method based on symmetry theory | |
| CN104463204A (en) | Target quantity statistical method | |
| CN104978569A (en) | Sparse representation based incremental face recognition method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130918 |
|
| RJ01 | Rejection of invention patent application after publication |