CN106383891A - Deep hash-based medical image distributed retrieval method - Google Patents
Deep hash-based medical image distributed retrieval method Download PDFInfo
- Publication number
- CN106383891A CN106383891A CN201610844011.7A CN201610844011A CN106383891A CN 106383891 A CN106383891 A CN 106383891A CN 201610844011 A CN201610844011 A CN 201610844011A CN 106383891 A CN106383891 A CN 106383891A
- Authority
- CN
- China
- Prior art keywords
- image
- hash
- output
- hadoop
- stride
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Library & Information Science (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a deep hash-based medical image distributed retrieval method. The method comprises deep hash extraction image features and Hadoop-based batch image feature matching parallel computing; in the deep hash extraction image features, similar or dissimilar image pairs are used as training input through a convolutional neural network model, the gradient of an objective function relative to multilayer network weights is calculated by use of a back propagation algorithm, and finally multiple output values of each image are guided to be approximate to discrete 0 or 1; and in the Hadoop-based batch image feature matching parallel computing, a feature file of batch images is divided into a plurality of blocks, the blocks are independent from each other, the blocks are distributed to different nodes for execution through an Apache Hadoop YARN (Yet Another Resource Negotiator) resource manager, and finally results executed by all Mappers are combined into a Reducer. According to the method provided by the invention, the gap between image representation and semanteme can be reduced, the retrieval precision is improved, the retrieval process is accelerated through parallel feature matching, and the retrieval efficiency of batch medical images is enhanced.
Description
Technical field
The present invention relates to medical Image Retrieval Technology field is and in particular to a kind of medical image based on depth Hash is distributed
Formula search method.
Background technology
At present, medical image system all creates increasing digitized image, such as X-ray in each medical domain
Figure, nuclear magnetic resonance, NMR figure, CT figure etc., these image great majority are stored in data base.Based on medical image retrieval, efficient group
Knit and manage these images to provide clinical diagnosises service, be the new model of medical diagnosiss.Medical image retrieval based on content
Mainly include two stages, medical image features extract and characteristic matching.
Image grey level histogram feature extraction be medical image features extract common method, the method with statistic Lai
The method of the histogram feature of reflection image mainly includes：
(1) average：Reflect the average gray of a width medical image.
Wherein, H (i) is the pixel number that pixel value is i.
(2) variance：Variance reflection be piece image gray scale Discrete Distribution situation numerically.
(3) gradient：Gradient reflection is the asymmetric degree that image histogram is distributed, skewness bigger expression Nogata
Figure distribution is more asymmetric, otherwise more symmetrical.
(4) kurtosis：Kurtosis reflection is intensity profile in the phase recency of draw value.
But, the present inventor finds through research, and existing algorithm has following defect in actual applications：
(1) locus residing for the local distribution of color and every kind of color in image cannot be described, lead to not describe in image
A certain specific object or object；(2) visual signature is usually the description to image statistics, only describes in terms of some
Characteristics of image, far from accurate expression piece image；(3) characteristic matching process, is image retrieval part the most time-consuming, works as figure
As library storage image very big when, single-unit point server can run into memory space and consume problem big and that retrieval rate is slow, do not have
Good autgmentability.
Content of the invention
The technical problem existing for artificial image's feature extracting method in prior art and single node image retrieval, this
A kind of medical image distributed search method based on depth Hash of bright offer, this search method can preferably allow characteristics of image
With vision or semantic similitude, thus improving the degree of accuracy of retrieval, and distributed storage and retrieval are realized by Hadoop framework.
In order to solve above-mentioned technical problem, present invention employs following technical scheme：
A kind of medical image distributed search method based on depth Hash, described distributed search method includes depth and breathes out
Wish and extract characteristics of image and calculated based on Hadoop batch images characteristic matching parallelization；Wherein,
Described depth Hash extracts characteristics of image and comprises the following steps：
S11, one convolutional neural networks model of design, this model includes the first convolutional layer, the first sampling that order is arranged
Layer, the second convolutional layer, the second sample level, the 3rd convolutional layer, the 3rd sample level, the first full articulamentum and the second full articulamentum；
S12, using similar or dissimilar image to as training input, through described convolutional neural networks model to image
Carry out multiple convolution layer, down-sampling layer, full articulamentum transmission, obtain overall cost function as follows：
Wherein, C is overall cost function, and N is image to logarithm, yiWhether i-th pair image similar, and 0 represent similar,
1 represents dissmilarity, ai,1It is the output result of first image in i-th pair image, ai,2It is second image in i-th pair image
Output result, and a=σ (z), z=wx+b, σ are ReLU activation primitive, and w is weight matrix, and x schemes for model outside input value
As pixel value, b is adjusting parameter, and θ is threshold value；
S13, according to overall cost function press minimization error method back propagation adjust weight matrix, until lose letter
Change in value amount is less than the threshold value of very little or reaches the iterationses specified, and training then terminates, before specially regarding formula (1) as
Latter two part seg1 and seg2, weights variable quantity is：
Therefore, the renewal formula of w is：
Wherein, η is learning rate,
S14, each image in image library is input to convolutional neural networks model designed by step S11 training
In, using output result as characteristics of image, and the vectorial binaryzation of output is encoded as Hash；
Described calculating based on Hadoop batch images characteristic matching parallelization is comprised the following steps：
S21, the batch images retrieving needs are input to the convolutional neural networks model designed by step S11 training
In, obtain characteristics of image file and upload in Hadoop, Hadoop can carry out piecemeal to characteristics of image file, and piecemeal is divided
To in different Mapper tasks it is assumed that characteristics of image file size is fileSize MB, each piecemeal mean size is
SplitSize MB, then have：
Wherein, the acquiescence block size of Hadoop2.X is 128MB, and n indicates n Mapper task；
Comprise feature and the Hash coding of image to be retrieved in S22, the input block of each Mapper, compiled according to Hash first
Code determines similar image Candidate Set, then accesses candidate image property data base, enters line retrieval and calculates special with image to be retrieved
Levy the similarity size of vector, that is, calculate the Euclidean distance of two characteristic vectors；
S23, the output result of all Mapper is merged in a Reducer, the set to each image to be retrieved,
According to similarity size, descending sort is carried out to the image of retrieval.
Further, described distributed search method also includes characteristics of image storage, and described image characteristic storage includes following
Step：Build Hash table and two tables of characteristics of image table respectively in MySQL, first pass through to train by each image in image library
Network model, using output result as characteristics of image, and using Image Name as row keyword, characteristics of image is stored in as content
In characteristics of image table, then using the result after characteristics of image binaryzation as row keyword, Image Name is stored in Hash as content
In table；Also Hash table and two tables of characteristics of image table are built respectively in HBase, by Sqoop by the data of two tables in MySQL
Import in the big table of HBase.
Further, in the convolutional neural networks model of design in described step S11, each layer parameter is described as follows：
First convolutional layer：Output quantity：32, convolution kernel size：11, stride：4, weight initialization type：xavier；
First sample level：Type：MAX, exports quantity：32, convolution kernel size：3, stride：2；
Second convolutional layer：Output quantity：32, convolution kernel size：5, stride：1, weight initialization type：xavier；
Second sample level：Type：AVE, exports quantity：32, convolution kernel size：3, stride：2；
3rd convolutional layer：Output quantity：64, convolution kernel size：5, stride：1, weight initialization type：xavier；
3rd sample level：Type：AVE, exports quantity：64, convolution kernel size：3, stride：2；
First full articulamentum：Output quantity：500；
Second full articulamentum：Output quantity：10.
Further, in described step S14, the vectorial binaryzation of output is specially：If the vector of output is more than 0, it is 1,
Otherwise it is then 0.
Compared with prior art, the medical image distributed search method based on depth Hash that the present invention provides, including
Depth Hash is extracted characteristics of image and is calculated based on Hadoop batch images characteristic matching parallelization, and it is special that depth Hash extracts image
Levying is by convolutional neural networks model, using similar or dissimilar image to as training input, using back-propagation algorithm
The gradient of calculating target function opposing layers network weight, the multiple output valve Approximation Discrete of final every image of guiding 0 or
1；Being calculated based on Hadoop batch images characteristic matching parallelization is that the tag file of batch images is divided into multiple pieces, block and block
Between be separate, these blocks pass through Apache Hadoop YARN (Yet Another Resource
Negotiator) explorer, is assigned to execution on different nodes, and the result after finally all Mapper have executed is all defeated
Enter in a Reducer, such retrieval result is unrelated with tag file piecemeal.Therefore, the distributed search method of the present invention
The wide gap between graphical representation and semanteme being reduced, thus improving retrieval accuracy, being added by parallelization characteristic matching simultaneously
Fast retrieving, strengthens the efficiency of batch medical image retrieval.
Brief description
Fig. 1 is the medical image distributed search method global design schematic diagram based on depth Hash that the present invention provides.
Fig. 2 is the convolutional neural networks CNN model structure schematic diagram that the present invention provides.
Fig. 3 is present invention offer based on Hadoop batch images characteristic matching parallelization computational methods schematic diagram.
Specific embodiment
In order that technological means, creation characteristic, reached purpose and effect that the present invention realizes are easy to understand, tie below
Conjunction is specifically illustrating, and the present invention is expanded on further.
In describing the invention it is to be understood that term " longitudinal ", " radially ", " length ", " width ", " thickness ",
" on ", D score, "front", "rear", "left", "right", " vertical ", " level ", " top ", " bottom ", " interior ", the orientation of instruction such as " outward " or
Position relationship is based on orientation shown in the drawings or position relationship, is for only for ease of the description present invention and simplifies description, and not
It is instruction or the hint device of indication or element must have specific orientation, with specific azimuth configuration and operation, therefore not
It is understood that as limitation of the present invention.In describing the invention, unless otherwise stated, " multiple " are meant that two or two
More than.
In describing the invention, it should be noted that unless otherwise clearly defined and limited, term " installation ", " phase
Even ", " connection " should be interpreted broadly, for example, it may be being fixedly connected or being detachably connected, or is integrally connected；Can
To be to be mechanically connected or electrical connection；Can be to be joined directly together it is also possible to be indirectly connected to by intermediary, Ke Yishi
The connection of two element internals.For the ordinary skill in the art, above-mentioned term can be understood at this with concrete condition
Concrete meaning in invention.
Refer to shown in Fig. 1-3, the present invention provides a kind of medical image distributed search method based on depth Hash, institute
State distributed search method to include depth Hash extraction characteristics of image and be based on Hadoop batch images characteristic matching parallelization meter
Calculate；Wherein,
Described depth Hash extracts characteristics of image and comprises the following steps：
S11, one convolutional neural networks model (Convolutional Neural Network, abbreviation CNN) of design, please
With reference to shown in Fig. 2, this model includes the first convolutional layer conv1, the first sample level pool1, the second convolutional layer that order arranges
Conv2, the second sample level pool2, the 3rd convolutional layer conv3, the 3rd sample level pool3, the first full articulamentum ip1 and second are complete
Articulamentum ip2, in described convolutional neural networks model, each layer parameter is described as follows：
First convolutional layer conv1：Output quantity：32, convolution kernel size：11, stride：4, weight initialization type：
xavier；
First sample level pool1：Type：MAX, exports quantity：32, convolution kernel size：3, stride：2；
Second convolutional layer conv2：Output quantity：32, convolution kernel size：5, stride：1, weight initialization type：
xavier；
Second sample level pool2：Type：AVE, exports quantity：32, convolution kernel size：3, stride：2；
3rd convolutional layer conv3：Output quantity：64, convolution kernel size：5, stride：1, weight initialization type：
xavier；
3rd sample level pool3：Type：AVE, exports quantity：64, convolution kernel size：3, stride：2；
First full articulamentum ip1：Output quantity：500, that is, the first full articulamentum ip1 comprise 500 nodes；
Second full articulamentum ip2：Output quantity：10, that is, the second full articulamentum ip2 comprise 10 nodes；
Using each layer parameter as described above, the vision process people simulating people by convolutional layer and sample level is external
The cognition on boundary is to overall from local, and the space relationship of image be also local pixel contact more tight, and distance is relatively
Remote pixel interdependence is then weaker.Thus, each neuron there is no need global image is perceived it is only necessary to play a game in fact
Portion is perceived, and then the informix of local gets up just obtained the information of the overall situation in higher.This method can be certainly
Feature in dynamic abstract image, the characteristics of image of higher-dimension is converted into the feature of low-dimensional, finally using full Connection Neural Network, this
Sample will reach good performance.
Choose cranium brain, breast, abdomen, spinal column and extremity 1000 respectively, each class random selection 80%, as training set, is left
20% as test set.
S12, propagated forward using similar or dissimilar image to as training input, image is from input layer (i.e. first
Convolutional layer conv1) through converting step by step, it is sent to output layer (i.e. the second full articulamentum ip2), this process is included through described
Convolutional neural networks model carries out multiple convolution layer, down-sampling layer, full articulamentum transmission to image, obtains overall cost function such as
Under：
Wherein, C is overall cost function, and N is image to logarithm, yiWhether i-th pair image similar, and 0 represent similar,
1 represents dissmilarity, ai,1It is the output result of first image in i-th pair image, ai,2It is second image in i-th pair image
Output result, and a=σ (z), z=wx+b, σ are ReLU activation primitive, and w is weight matrix, and x schemes for model outside input value
As pixel value, b is adjusting parameter, and θ is threshold value.
S13, back-propagation：Often can be damaged after convolutional neural networks model designed by through step S11 for a collection of image
Lose functional value, weight matrix is adjusted according to the method back propagation that overall cost function presses minimization error, until losing letter
Change in value amount is less than the threshold value of very little or reaches the iterationses specified, and training then terminates.Formula (1) is regarded as former and later two
Part seg1 and seg2, weights variable quantity is：
Therefore, the renewal formula of w is：
Wherein, η is learning rate,
Carry out parameter optimization, network parameter including learning rate, iterationses etc. for the adjustment, and the knot to network output
Fruit is counted, and feels out proper parameter as optimal value.
S14, each image in image library is input to convolutional neural networks model designed by step S11 training
In, using output result as characteristics of image, and the vectorial binaryzation of output is encoded as Hash, the described vector two by output
Value is specially：If the vector of output is more than 0, it is 1, otherwise be then 0.
Refer to shown in Fig. 1, build two tables, respectively Hash table and characteristics of image table in MySQL, first by image library
Each image passes through the network model training, and using output result as characteristics of image, and using Image Name as row keyword, schemes
As feature is stored in characteristics of image table as content, its concrete storage form is：<imageidi, feature>；Again image
As row keyword (i.e. Hash coding), Image Name is stored in Hash table result after feature binaryzation as content, its tool
Body storage form is：<Hashcode, imageid1imageid2...imageidk>, wherein k is Hash code length, that is, second
The node number that full articulamentum ip2 is comprised, in the foregoing embodiment k be configured to 10；Meanwhile, two are also built in HBase
Open table, respectively Hash table and characteristics of image table, by Sqoop, the data of two tables in MySQL is imported to the big table of HBase
In, thus can ensure that the data in HBase can constantly expand.
Refer to shown in Fig. 3, described calculating based on Hadoop batch images characteristic matching parallelization is comprised the following steps：
S21, the batch images retrieving needs are input to the convolutional neural networks model designed by step S11 training
In, obtain characteristics of image file and upload in Hadoop, Hadoop can carry out piecemeal to characteristics of image file, and piecemeal is divided
To in different Mapper tasks it is assumed that characteristics of image file size is fileSize MB, each piecemeal mean size is
SplitSize MB, then have：
Wherein, the acquiescence block size of Hadoop2.X is 128MB, and n indicates n Mapper task, and formula (5) is meant to be
Average piecemeal, each task obtains a piecemeal, and consuming total time that therefore whole cluster completes to search for is minimum；
As a kind of specific embodiment, the form of described image tag file is：mageid_hashcode_
Features, wherein, imageid is image recognition unique identifier (image name), and hashcode is that the index of this image is compiled
Number, features is the characteristic vector of this image.
Comprise feature and the Hash coding of image to be retrieved in S22, the input block of each Mapper, compiled according to Hash first
Code hashcode is indexed in HBase Hash table, obtains similar image Candidate Set, then accesses candidate image characteristic
Storehouse, enters line retrieval and calculates similarity size dist with image feature vector to be retrieved, that is, calculate the Europe of two characteristic vectors
Formula distance；And dist is bigger, represent that two characteristic vector distances are more remote, that is, similarity is lower, and in the same manner, dist is less, and similarity is got over
High；
S23, the output result of all Mapper is merged (combine) in a Reducer, to each figure to be retrieved
The set of picture, carries out descending sort according to similarity size to the image of retrieval, you can with according to dist size (the less phase of dist
Higher like degree) similar image (similar imageid) is carried out by ascending sort and exports retrieval result, for example can show front 20
Open most like image.
Wherein, Hadoop Distributed Computing Platform includes most crucial distributed file system HDFS (Hadoop
Distributed file system), MapReduce and HBase, HDFS be Hadoop use a distributed file
System, HBase is a distributed NoSQL data base, and MapReduce is a kind of simple but powerful programming model,
Can parallel processing large data collection.And the operation of MapReduce comprises two steps：After the complete input data of Map phase process, defeated
Go out<Key, value>Key-value pair；In the Reduce stage, collect and process key assignments identical<Key, value>Key-value pair.
Compared with prior art, the medical image distributed search method based on depth Hash that the present invention provides, including
Depth Hash is extracted characteristics of image and is calculated based on Hadoop batch images characteristic matching parallelization, and it is special that depth Hash extracts image
Levying is by convolutional neural networks model, using similar or dissimilar image to as training input, using back-propagation algorithm
The gradient of calculating target function opposing layers network weight, the multiple output valve Approximation Discrete of final every image of guiding 0 or
1；Being calculated based on Hadoop batch images characteristic matching parallelization is that the tag file of batch images is divided into multiple pieces, block and block
Between be separate, these blocks pass through Apache Hadoop YARN (Yet Another Resource
Negotiator) explorer, is assigned to execution on different nodes, and the result after finally all Mapper have executed is all defeated
Enter in a Reducer, such retrieval result is unrelated with tag file piecemeal.Therefore, the distributed search method of the present invention
The wide gap between graphical representation and semanteme being reduced, thus improving retrieval accuracy, being added by parallelization characteristic matching simultaneously
Fast retrieving, strengthens the efficiency of batch medical image retrieval.
Finally illustrate, above example only in order to technical scheme to be described and unrestricted, although with reference to relatively
Good embodiment has been described in detail to the present invention, it will be understood by those within the art that, can be to the skill of the present invention
Art scheme is modified or equivalent, the objective without deviating from technical solution of the present invention and scope, and it all should be covered at this
In the middle of the right of invention.
Claims (4)
1. a kind of medical image distributed search method based on depth Hash is it is characterised in that described distributed search method
Extract characteristics of image including depth Hash and based on Hadoop batch images characteristic matching parallelization calculating；Wherein,
Described depth Hash extracts characteristics of image and comprises the following steps：
S11, one convolutional neural networks model of design, this model includes the first convolutional layer, the first sample level, that order arranges
Two convolutional layers, the second sample level, the 3rd convolutional layer, the 3rd sample level, the first full articulamentum and the second full articulamentum；
S12, using similar or dissimilar image to as training input, through described convolutional neural networks model, image is carried out
Multiple convolution layer, down-sampling layer, full articulamentum transmission, obtain overall cost function as follows：
Wherein, C is overall cost function, and N is image to logarithm, yiIt is whether i-th pair image is similar, and 0 represents similar, 1 representative
Dissmilarity, ai,1It is the output result of first image in i-th pair image, ai,2It is the output of second image in i-th pair image
As a result, and a=σ (z), z=wx+b, σ are ReLU activation primitive, and w is weight matrix, and x is image slices for model outside input value
Element value, b is adjusting parameter, and θ is threshold value；
S13, according to overall cost function press minimization error method back propagation adjust weight matrix, until loss function value
Variable quantity is less than the threshold value of very little or reaches the iterationses specified, and training then terminates, and specially regards formula (1) as before and after two
Individual part seg1 and seg2, weights variable quantity is：
Therefore, the renewal formula of w is：
Wherein, η is learning rate,
S14, each image in image library is input in the convolutional neural networks model designed by step S11 training,
Using output result as characteristics of image, and the vectorial binaryzation of output is encoded as Hash；
Described calculating based on Hadoop batch images characteristic matching parallelization is comprised the following steps：
S21, the batch images retrieving needs are input in the convolutional neural networks model designed by step S11 training,
Obtain characteristics of image file and upload in Hadoop, Hadoop can carry out piecemeal to characteristics of image file, and piecemeal is assigned to
It is assumed that characteristics of image file size is fileSize MB in different Mapper tasks, each piecemeal mean size is
SplitSize MB, then have：
Wherein, the acquiescence block size of Hadoop2.X is 128MB, and n indicates n Mapper task；
Feature and the Hash coding of image to be retrieved is comprised, first according to Hash coding really in S22, the input block of each Mapper
Determine similar image Candidate Set, then access candidate image property data base, enter line retrieval and calculate with characteristics of image to be retrieved to
The similarity size of amount, that is, calculate the Euclidean distance of two characteristic vectors；
S23, the output result of all Mapper is merged in a Reducer, the set to each image to be retrieved, according to
Similarity carries out descending sort to the image of retrieval greatly.
2. the medical image distributed search method based on depth Hash according to claim 1 is it is characterised in that described
Distributed search method also includes characteristics of image storage, and described image characteristic storage comprises the following steps：Build Kazakhstan in MySQL respectively
Each image in image library is first passed through the network model training, by output result by uncommon table and two tables of characteristics of image table
As characteristics of image, and using Image Name as row keyword, characteristics of image is stored in characteristics of image table as content, then figure
As the result after feature binaryzation is as row keyword, Image Name is stored in Hash table as content；HBase also distinguishes
Build Hash table and two tables of characteristics of image table, by Sqoop, the data of two tables in MySQL is imported to the big table of HBase
In.
3. the medical image distributed search method based on depth Hash according to claim 1 is it is characterised in that described
In the convolutional neural networks model of design in step S11, each layer parameter is described as follows：
First convolutional layer：Output quantity：32, convolution kernel size：11, stride：4, weight initialization type：xavier；
First sample level：Type：MAX, exports quantity：32, convolution kernel size：3, stride：2；
Second convolutional layer：Output quantity：32, convolution kernel size：5, stride：1, weight initialization type：xavier；
Second sample level：Type：AVE, exports quantity：32, convolution kernel size：3, stride：2；
3rd convolutional layer：Output quantity：64, convolution kernel size：5, stride：1, weight initialization type：xavier；
3rd sample level：Type：AVE, exports quantity：64, convolution kernel size：3, stride：2；
First full articulamentum：Output quantity：500；
Second full articulamentum：Output quantity：10.
4. the medical image distributed search method based on depth Hash according to claim 1 is it is characterised in that described
In step S14, the vectorial binaryzation of output is specially：If the vector of output is more than 0, it is 1, otherwise be then 0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610844011.7A CN106383891B (en) | 2016-09-22 | 2016-09-22 | A kind of medical image distributed search method based on depth Hash |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610844011.7A CN106383891B (en) | 2016-09-22 | 2016-09-22 | A kind of medical image distributed search method based on depth Hash |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106383891A true CN106383891A (en) | 2017-02-08 |
| CN106383891B CN106383891B (en) | 2019-04-23 |
Family
ID=57936747
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610844011.7A Expired - Fee Related CN106383891B (en) | 2016-09-22 | 2016-09-22 | A kind of medical image distributed search method based on depth Hash |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106383891B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107092918A (en) * | 2017-03-29 | 2017-08-25 | 太原理工大学 | It is a kind of to realize that Lung neoplasm sign knows method for distinguishing based on semantic feature and the image retrieval for having supervision Hash |
| CN107315765A (en) * | 2017-05-12 | 2017-11-03 | 南京邮电大学 | A kind of method of the concentrated-distributed proximity search of extensive picture |
| CN107402947A (en) * | 2017-03-29 | 2017-11-28 | 北京粉笔未来科技有限公司 | Picture retrieval method for establishing model and device, picture retrieval method and device |
| CN107688823A (en) * | 2017-07-20 | 2018-02-13 | 北京三快在线科技有限公司 | A kind of characteristics of image acquisition methods and device, electronic equipment |
| CN107992573A (en) * | 2017-11-30 | 2018-05-04 | 公安部第三研究所 | Distributed vector index method and system based on position sensing Hash |
| CN108932314A (en) * | 2018-06-21 | 2018-12-04 | 南京农业大学 | A kind of chrysanthemum image content retrieval method based on the study of depth Hash |
| CN109828953A (en) * | 2019-01-30 | 2019-05-31 | 武汉虹旭信息技术有限责任公司 | Picture retrieval system and its method based on distributed memory database |
| CN109871461A (en) * | 2019-02-13 | 2019-06-11 | 华南理工大学 | The large-scale image sub-block search method to be reordered based on depth Hash network and sub-block |
| CN109901164A (en) * | 2019-03-21 | 2019-06-18 | 桂林电子科技大学 | A kind of distributed rear orientation projection's imaging method of synthetic aperture radar |
| CN110796182A (en) * | 2019-10-15 | 2020-02-14 | 西安网算数据科技有限公司 | Bill classification method and system for small amount of samples |
| CN110998607A (en) * | 2017-08-08 | 2020-04-10 | 三星电子株式会社 | System and method for neural networks |
| CN111160535A (en) * | 2019-12-31 | 2020-05-15 | 北京计算机技术及应用研究所 | DGCNN model acceleration method based on Hadoop |
| WO2020125100A1 (en) * | 2018-12-21 | 2020-06-25 | 华为技术有限公司 | Image search method, apparatus, and device |
| CN111581420A (en) * | 2020-04-30 | 2020-08-25 | 徐州医科大学 | Medical image real-time retrieval method based on Flink |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104834748A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Image retrieval method utilizing deep semantic to rank hash codes |
| US20150242463A1 (en) * | 2014-02-25 | 2015-08-27 | Tsung-Han Lin | Systems, apparatuses, and methods for deep learning of feature detectors with sparse coding |
| CN105354307A (en) * | 2015-11-06 | 2016-02-24 | 腾讯科技（深圳）有限公司 | Image content identification method and apparatus |
| US20160098844A1 (en) * | 2014-10-03 | 2016-04-07 | EyeEm Mobile GmbH | Systems, methods, and computer program products for searching and sorting images by aesthetic quality |
| CN105550222A (en) * | 2015-12-07 | 2016-05-04 | 中国电子科技网络信息安全有限公司 | Distributed storage-based image service system and method |
-
2016
- 2016-09-22 CN CN201610844011.7A patent/CN106383891B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150242463A1 (en) * | 2014-02-25 | 2015-08-27 | Tsung-Han Lin | Systems, apparatuses, and methods for deep learning of feature detectors with sparse coding |
| US20160098844A1 (en) * | 2014-10-03 | 2016-04-07 | EyeEm Mobile GmbH | Systems, methods, and computer program products for searching and sorting images by aesthetic quality |
| CN104834748A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Image retrieval method utilizing deep semantic to rank hash codes |
| CN105354307A (en) * | 2015-11-06 | 2016-02-24 | 腾讯科技（深圳）有限公司 | Image content identification method and apparatus |
| CN105550222A (en) * | 2015-12-07 | 2016-05-04 | 中国电子科技网络信息安全有限公司 | Distributed storage-based image service system and method |
Non-Patent Citations (2)
| Title |
|---|
| 彭天强等: "基于深度卷积神经网络和二进制哈希学习的图像检索方法", 《电子与信息学报》 * |
| 龚震霆等: "基于卷积神经网络和哈希编码的图像检索方法", 《智能系统学报》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107092918A (en) * | 2017-03-29 | 2017-08-25 | 太原理工大学 | It is a kind of to realize that Lung neoplasm sign knows method for distinguishing based on semantic feature and the image retrieval for having supervision Hash |
| CN107402947A (en) * | 2017-03-29 | 2017-11-28 | 北京粉笔未来科技有限公司 | Picture retrieval method for establishing model and device, picture retrieval method and device |
| CN107402947B (en) * | 2017-03-29 | 2020-12-08 | 北京猿力教育科技有限公司 | Picture retrieval model establishing method and device and picture retrieval method and device |
| CN107092918B (en) * | 2017-03-29 | 2020-10-30 | 太原理工大学 | Image retrieval method based on semantic features and supervised hashing |
| CN107315765A (en) * | 2017-05-12 | 2017-11-03 | 南京邮电大学 | A kind of method of the concentrated-distributed proximity search of extensive picture |
| CN107688823A (en) * | 2017-07-20 | 2018-02-13 | 北京三快在线科技有限公司 | A kind of characteristics of image acquisition methods and device, electronic equipment |
| US11282295B2 (en) | 2017-07-20 | 2022-03-22 | Beijing Sankuai Online Technology Co., Ltd | Image feature acquisition |
| CN110998607A (en) * | 2017-08-08 | 2020-04-10 | 三星电子株式会社 | System and method for neural networks |
| CN110998607B (en) * | 2017-08-08 | 2024-03-08 | 三星电子株式会社 | System and method for neural networks |
| CN107992573A (en) * | 2017-11-30 | 2018-05-04 | 公安部第三研究所 | Distributed vector index method and system based on position sensing Hash |
| CN108932314A (en) * | 2018-06-21 | 2018-12-04 | 南京农业大学 | A kind of chrysanthemum image content retrieval method based on the study of depth Hash |
| WO2020125100A1 (en) * | 2018-12-21 | 2020-06-25 | 华为技术有限公司 | Image search method, apparatus, and device |
| CN109828953A (en) * | 2019-01-30 | 2019-05-31 | 武汉虹旭信息技术有限责任公司 | Picture retrieval system and its method based on distributed memory database |
| CN109871461A (en) * | 2019-02-13 | 2019-06-11 | 华南理工大学 | The large-scale image sub-block search method to be reordered based on depth Hash network and sub-block |
| CN109871461B (en) * | 2019-02-13 | 2020-12-22 | 华南理工大学 | Large-scale image subblock retrieval method based on deep hash network and subblock reordering |
| CN109901164A (en) * | 2019-03-21 | 2019-06-18 | 桂林电子科技大学 | A kind of distributed rear orientation projection's imaging method of synthetic aperture radar |
| CN110796182A (en) * | 2019-10-15 | 2020-02-14 | 西安网算数据科技有限公司 | Bill classification method and system for small amount of samples |
| CN111160535A (en) * | 2019-12-31 | 2020-05-15 | 北京计算机技术及应用研究所 | DGCNN model acceleration method based on Hadoop |
| CN111160535B (en) * | 2019-12-31 | 2024-01-30 | 北京计算机技术及应用研究所 | DGCNN model acceleration method based on Hadoop |
| CN111581420A (en) * | 2020-04-30 | 2020-08-25 | 徐州医科大学 | Medical image real-time retrieval method based on Flink |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106383891B (en) | 2019-04-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106383891A (en) | Deep hash-based medical image distributed retrieval method | |
| Wu et al. | Object detection based on RGC mask R‐CNN | |
| US20210042580A1 (en) | Model training method and apparatus for image recognition, network device, and storage medium | |
| CN110309856A (en) | Image classification method, the training method of neural network and device | |
| CN109920501A (en) | Electronic health record classification method and system based on convolutional neural networks and Active Learning | |
| CN110490239B (en) | Training method, quality classification method, device and equipment of image quality control network | |
| CN110070107A (en) | Object identification method and device | |
| CN110059198A (en) | A kind of discrete Hash search method across modal data kept based on similitude | |
| CN106844524B (en) | A kind of medical image search method converted based on deep learning and Radon | |
| Ahmad et al. | Genetic algorithm-artificial neural network (GA-ANN) hybrid intelligence for cancer diagnosis | |
| CN109948149A (en) | A kind of file classification method and device | |
| CN104112143A (en) | Weighted hyper-sphere support vector machine algorithm based image classification method | |
| WO2021051987A1 (en) | Method and apparatus for training neural network model | |
| CN108416364A (en) | Integrated study data classification method is merged in subpackage | |
| CN106682681A (en) | Recognition algorithm automatic improvement method based on relevance feedback | |
| CN110309835A (en) | A kind of image local feature extracting method and device | |
| CN113569895A (en) | Image processing model training method, processing method, device, equipment and medium | |
| CN113538359B (en) | System and method for finger vein image segmentation | |
| CN113705276A (en) | Model construction method, model construction device, computer apparatus, and medium | |
| CN110335160A (en) | A kind of medical treatment migratory behaviour prediction technique and system for improving Bi-GRU based on grouping and attention | |
| Perveen et al. | Multidimensional Attention-Based CNN Model for Identifying Apple Leaf Disease. | |
| CN109344309A (en) | Extensive file and picture classification method and system are stacked based on convolutional neural networks | |
| CN116109868A (en) | Image classification model construction and small sample image classification method based on lightweight neural network | |
| Zhao et al. | Corner‐based object detection method for reactivating box constraints | |
| WO2021114626A1 (en) | Method for detecting quality of medical record data and related device |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190423 Termination date: 20210922 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |