CN103268494B - Parasite egg recognition methods based on rarefaction representation - Google Patents
Parasite egg recognition methods based on rarefaction representation Download PDFInfo
- Publication number
- CN103268494B CN103268494B CN201310181012.4A CN201310181012A CN103268494B CN 103268494 B CN103268494 B CN 103268494B CN 201310181012 A CN201310181012 A CN 201310181012A CN 103268494 B CN103268494 B CN 103268494B
- Authority
- CN
- China
- Prior art keywords
- dictionary
- identification
- rarefaction representation
- parasite egg
- initial
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to image identification technical field, be specifically related to a kind of parasite egg recognition methods based on rarefaction representation, including: set up initial dictionary; Use K-SVD algorithm that dictionary is learnt; Process input picture; Calculate reconstruction error matrix; Obtain candidate image block; Identify the steps such as candidate image block. Invention introduces the sorting algorithm based on rarefaction representation, enhance the whole parasite egg recognizer robustness to various interference factors; Introduce Batch-OMP algorithm and represent process for Large Scale Sparse, improve recognition efficiency; Introduce the method that dictionary directly set up by the sample after with gaussian pyramid dimensionality reduction, it is to avoid extract the step of worm's ovum target characteristic, make the identification process become more easy; Introduce the method set up error matrix and seek its local minimum, it is to avoid in Primary Location process, obtain comprising the different images block of same target.
Description
Technical field
The invention belongs to image identification technical field, be specifically related to a kind of parasite egg recognition methods based on rarefaction representation.
Background technology
Be based upon that the parasite egg on Computer Image Processing and medical science microtechnique identifies automatically it is crucial that design image recognition algorithm fast and effectively, in the past based on the parasite egg automatic identifying method of image mainly by means of first isolating worm's ovum target, extract the various features of target again, finally complete to identify in conjunction with a grader. Lift the method that two examples are more relevant to the present invention: 16 kinds of parasite of human worm's ovums are not identified by bending moment and support vector machine in document " Anexpertdiagnosissystemforclassificationofhumanparasitee ggsbasedonmulti-classSVM " in conjunction with 7 of Hu for (1) DeryaAvci et al. 2009, though obtaining significantly high discrimination, but only can be only achieved under the premise that image is ideal, do not consider situation when interference factor is more; (2) Chinese patent CN201110022426.3 proposes a kind of method in conjunction with parasite egg edge histogram parasite of human worm's ovum is carried out shape recognition, overcome the impact of weak boundary preferably, improve the reliability of identification, but for the comparatively similar parasitic shape recognition of shape or Shortcomings. From existing method, the kind of feature is more, also including color, shape, size, texture etc. except the feature described in said method, feature selection obtains the fine or not last discrimination that largely determines, simultaneously Primary Location target extract that the step of feature is also more difficult to be completed accurately. Grader is also varied, including Bayes classifier, linear discriminant analysis, support vector machine, neutral net, minimum range etc., because these graders are to being characterized by sensitivity, then selecting which kind of feature is that optimum being often difficult to is determined for grader, simultaneously these graders to such as noise, block, the robustness of the interference factor such as impurity all more weak.
Application based on rarefaction representation sorting algorithm also launches far away, such a rudimentary algorithm framework is in different application scenarios, need in conjunction with other technologies and skill algorithm to be transformed and expand, need to determine according to specific needs in Data Dimensionality Reduction, the selection of rarefaction representation algorithm, dictionary learning especially.Based on above analysis, it is applied in parasite egg identification problem first, it is achieved the identification of single class or multiclass parasite egg.
Summary of the invention
It is an object of the invention to overcome conventional parasite egg recognition methods to feature and the more sensitive defect of various interference factor, in conjunction with being suitable for Batch-OMP algorithm and the K-SVD dictionary learning algorithm that Large Scale Sparse represents, propose a kind of parasite egg recognition methods based on rarefaction representation, with the application demand of satisfied actual parasite egg automatic recognition system on discrimination and recognition efficiency.
In order to realize foregoing invention purpose, the present invention by the following technical solutions: a kind of parasite egg recognition methods based on rarefaction representation, comprise the following steps:
(1) initial dictionary is set up: initial single category dictionary is set up in single class identification, and initial joint dictionary is set up in multiclass identification;
(2) dictionary learning: use K-SVD algorithm that dictionary is learnt, single class identification obtains single class and represents that dictionary, multiclass identification obtain combining expression dictionary and combining differentiation dictionary;
(3) processing input picture: input picture is carried out pyramidal compression, by the mode of sliding window, compression image is carried out piecemeal, step-length can be chosen as one or more pixel; Based on Batch-OMP algorithm, all image blocks are carried out rarefaction representation, and the dictionary of single class identification adopts single class to represent dictionary, and the dictionary of multiclass identification adopts associating expression dictionary;
(4) reconstruction error matrix is calculated;
(5) obtain candidate image block: utilize the reconstruction error matrix obtained in step (4), find its all of local minimum, choose the corresponding image block of k wherein minimum value as candidate target;
(6) candidate image block is identified: for single class identification situation, differentiating candidate image block by threshold value, identification completes; For multiclass identification situation, candidate image block carrying out rarefaction representation, use associating differentiation dictionary, calculate sub-dictionary reconstruction error, use threshold mode to carry out differentiating to candidate image block and classify, identification completes.
In step (1), set up initial dictionary step as follows:
(1) the parasite egg pictures sample c n selecting some impurity less and representative is individual, and wherein c is the integer of >=1, represents class number, and n represents the sample number of each class;
(2) adopt gaussian pyramid that c n width image is compressed, obtain the image pattern after dimensionality reduction;
(3) each image obtained in previous step is rotated a circle with d degree for step-length obtaining including 360/d image pattern of artwork, then total number of samples is N=360 c n/d;
(4) it is one-dimensional vector by each two-dimensional image data " elongation " obtained in the previous step, then is standardized each vector processing, make each vector meet l2-norm is 1;
(5) using all standardized vectors obtained in the previous step as the atom of dictionary, initial dictionary is obtained, if c=1, what then obtain is initial single category dictionary of single class identification, if c > 1, then what obtain is the initial joint dictionary of multiclass identification, comprises c sub-dictionary.
In step (2), use K-SVD algorithm that dictionary is learnt, be divided into three kinds of situations:
(1) for single class parasite egg identification, with K-SVD algorithm, initial single category dictionary being learnt, obtain single class and represent dictionary, this dictionary is simultaneously used for Primary Location and classification, and the volume of dictionary is determined according to the dimension of atom vector;
(2) for multiclass parasite egg identification, with K-SVD algorithm, whole initial joint dictionary being learnt, obtain associating expression dictionary, this dictionary is used for Primary Location;
(3) for multiclass parasite egg identification, with K-SVD algorithm, each initial sub-dictionary is learnt, being combined by sub-dictionary after all study and obtain associating differentiation dictionary, this dictionary is used for classifying, and its volume represents the volume of dictionary much larger than combining.
In step (3), it is that Large Scale Sparse represents that all image blocks carry out rarefaction representation, namely uses Batch-OMP Algorithm for Solving formula (1-1):
min||x-Dθ||2s.t.||θ||0≤T(1-1)
Wherein x is input signal, and D is that the single class obtained in step (2) represents dictionary or combines expression dictionary, and θ is coefficient, and T is openness condition.
In step (4), calculating reconstruction error matrix, step is as follows:
(1) utilize formula (1-2) to calculate reconstruction error, obtain the reconstruction error [e of all image blocks1,e2,…,eL], wherein L is image block number;
Wherein x is input signal, and D represents dictionary for single class or combines expression dictionary, and θ is coefficient.
(2) order of the image block obtained according to step (3), by [e1,e2,…,eL] arranged in sequence becomes a two-dimensional matrix, what obtain is reconstruction error matrix.
In step (6), calculate sub-dictionary reconstruction error, calculate according to formula (1-3).
Wherein x is input signal, DiFor associating dictionary D=[D1,D2,…,Dc] sub-dictionary, wherein i=1,2 ..., c, c is class number, and θ is coefficient.
The parasite egg recognition methods based on rarefaction representation of the present invention, introduces the sorting algorithm based on rarefaction representation, enhances the whole parasite egg recognizer robustness to various interference factors; Introduce Batch-OMP algorithm and represent process for Large Scale Sparse, improve recognition efficiency; Introduce the method that dictionary directly set up by the sample after with gaussian pyramid dimensionality reduction, it is to avoid extract the step of worm's ovum target characteristic, make the identification process become more easy; Introduce the method set up error matrix and seek its local minimum, it is to avoid in Primary Location process, obtain comprising the different images block of same target.
Accompanying drawing explanation
Fig. 1 is the present invention flow chart based on the parasite egg recognition methods of rarefaction representation.
A few class parasite egg samples that Fig. 2 chooses.
Fig. 3 is single category dictionary initially.
Fig. 4 initial joint dictionary.
The mono-class of Fig. 5 represents dictionary.
Fig. 6 combines expression dictionary.
Fig. 7 combines differentiation dictionary.
Fig. 8 input picture.
Fig. 9 reconstruction error vector.
Figure 10 reconstruction error matrix.
Detailed description of the invention
Below in conjunction with specific embodiment, the present invention will be further explained.
As it is shown in figure 1, be the flow process of the parasite egg recognition methods based on rarefaction representation of the present invention, comprise the following steps:
(1) initial dictionary is set up: initial single category dictionary is set up in single class identification, and initial joint dictionary is set up in multiclass identification. The first step: the parasite egg pictures sample c n selecting some impurity less and representative is individual, wherein c >=1 represents class number, and n represents the sample number of each class, as shown in Figure 2. Second step: adopt gaussian pyramid that c n width image is compressed, obtain the image pattern after dimensionality reduction. 3rd step: each image obtained in previous step being rotated a circle with d degree for step-length and obtains including 360/d image pattern of artwork, then total number of samples is N=360 c n/d, it is possible to take d=5. 4th step: be one-dimensional vector by each two-dimensional image data " elongation " obtained in the previous step, then be standardized each vector processing, make each vector meet l2-norm is 1. 5th step: using all standardized vectors obtained in the previous step atom as dictionary, obtain initial dictionary, if c=1, what then obtain is initial single category dictionary, if as it is shown on figure 3, c > 1, what then obtain is the initial joint dictionary of multiclass, comprises c sub-dictionary, as shown in Figure 4.
(2) dictionary learning: use K-SVD algorithm that dictionary is learnt, single class identification obtains single class and represents that dictionary, multiclass identification obtain combining expression dictionary and combining differentiation dictionary, is specifically divided into three kinds of situations. The first: for single class parasite egg identification, with K-SVD algorithm, initial single category dictionary being learnt, obtain single class and represent dictionary, as it is shown in figure 5, this dictionary is simultaneously used for Primary Location and classification, the volume of dictionary is determined according to the dimension of atom vector. The second: for multiclass parasite egg identification, learns whole initial joint dictionary with K-SVD algorithm, obtains associating expression dictionary, and as shown in Figure 6, this dictionary is used for Primary Location. The third: is for multiclass parasite egg identification, with K-SVD algorithm, each initial sub-dictionary is learnt, the more sub-dictionary after all study is combined obtain associating differentiation dictionary, as shown in Figure 7, this dictionary is used for classifying, and its volume represents the volume of dictionary much larger than combining.
(3) processing input picture: input picture is carried out pyramidal compression, by the mode of sliding window, compression image is carried out piecemeal, step-length can be chosen as one or more pixel; Based on Batch-OMP algorithm, all image blocks are carried out rarefaction representation, and the dictionary of single class identification adopts single class to represent dictionary, and the dictionary of multiclass identification adopts associating expression dictionary.
In step (3), all image blocks are carried out rarefaction representation and relates to Large Scale Sparse and represent, use Batch-OMP Algorithm for Solving formula (1-1).
min||x-Dθ||2s.t.||θ||0≤T(1-1)
Wherein x is input signal, and D is that the single class obtained in step (2) represents dictionary or combines expression dictionary, and θ is coefficient, and T is openness condition.
(4) reconstruction error matrix is calculated. The first step: utilize formula (1-2) to calculate reconstruction error, obtain the reconstruction error [e of all image blocks1,e2,…,eL], as shown in Fig. 8 (input picture) and Fig. 9 (reconstruction error vector).
Wherein x is input signal, and D represents dictionary for single class or combines expression dictionary, and θ is coefficient.
Second step: the order according to the image block that step (3) obtains, by [e1,e2,…,eL] arranged in sequence becomes a two-dimensional matrix, what obtain is reconstruction error matrix, as shown in Figure 10.
(5) candidate image block is obtained: utilize the reconstruction error matrix obtained in step (4), find its all of local minimum, choose the corresponding image block of k wherein minimum value as candidate target, the local minimum point distribution in Figure 10.
(6) candidate image block is identified: for single class identification situation, differentiating candidate image block by threshold value, identification completes; For multiclass identification situation, candidate image block is carried out rarefaction representation, differentiation dictionary is combined in use, calculate sub-dictionary reconstruction error, use threshold mode to carry out differentiating to candidate target and classify, the candidate image block of threshold value both is greater than for reconstruction error and is judged as without known class target, for the reconstruction error candidate image block less than threshold value, its classification belongs to the class corresponding to sub-dictionary that reconstruction error is minimum, and identification completes.
Calculate sub-dictionary reconstruction error in this step, calculate according to formula (1-3).
Wherein DiFor associating dictionary D=[D1,D2,…,Dc] sub-dictionary, wherein i=1,2 ..., c, c is class number.
Claims (6)
1. the parasite egg recognition methods based on rarefaction representation, it is characterised in that comprise the following steps:
(1) initial dictionary is set up: initial single category dictionary is set up in single class identification, and initial joint dictionary is set up in multiclass identification;
(2) dictionary learning: use K-SVD algorithm that dictionary is learnt, single class identification obtains single class and represents that dictionary, multiclass identification obtain combining expression dictionary and combining differentiation dictionary;
(3) processing input picture: input picture is carried out pyramidal compression, by the mode of sliding window, compression image is carried out piecemeal, step-length can be chosen as one or more pixel; Based on Batch-OMP algorithm, all image blocks are carried out rarefaction representation, and the dictionary of single class identification adopts single class to represent dictionary, and the dictionary of multiclass identification adopts associating expression dictionary;
(4) reconstruction error matrix is calculated;
(5) obtain candidate image block: utilize the reconstruction error matrix obtained in step (4), find its all of local minimum, choose the corresponding image block of k wherein minimum value as candidate target;
(6) candidate image block is identified: for single class identification situation, differentiating candidate image block by threshold value, identification completes; For multiclass identification situation, candidate image block carrying out rarefaction representation, use associating differentiation dictionary, calculate sub-dictionary reconstruction error, use threshold mode to carry out differentiating to candidate image block and classify, identification completes.
2. the parasite egg recognition methods based on rarefaction representation according to claim 1, it is characterised in that: in step (1), described foundation initial dictionary step is as follows:
(1) the parasite egg pictures sample c n selecting some impurity less and representative is individual, and wherein c is the integer of >=1, represents class number, and n represents the sample number of each class;
(2) adopt gaussian pyramid that c n width image is compressed, obtain the image pattern after dimensionality reduction;
(3) each image obtained in previous step is rotated a circle with d degree for step-length obtaining including 360/d image pattern of artwork, then total number of samples is N=360 c n/d;
(4) it is one-dimensional vector by each two-dimensional image data " elongation " obtained in the previous step, then is standardized each vector processing, make each vector meet l2-norm is 1;
(5) using all standardized vectors obtained in the previous step as the atom of dictionary, initial dictionary is obtained, if c=1, what then obtain is initial single category dictionary of single class identification, if c > 1, then what obtain is the initial joint dictionary of multiclass identification, comprises c sub-dictionary.
3. the parasite egg recognition methods based on rarefaction representation according to claim 1, it is characterised in that: in step (2), dictionary is learnt by described use K-SVD algorithm, is divided into three kinds of situations:
(1) for single class parasite egg identification, with K-SVD algorithm, initial single category dictionary being learnt, obtain single class and represent dictionary, this dictionary is simultaneously used for Primary Location and classification, and the volume of dictionary is determined according to the dimension of atom vector;
(2) for multiclass parasite egg identification, with K-SVD algorithm, whole initial joint dictionary being learnt, obtain associating expression dictionary, this dictionary is used for Primary Location;
(3) for multiclass parasite egg identification, with K-SVD algorithm, each initial sub-dictionary is learnt, being combined by sub-dictionary after all study and obtain associating differentiation dictionary, this dictionary is used for classifying, and its volume represents the volume of dictionary much larger than combining.
4. the parasite egg recognition methods based on rarefaction representation according to claim 1, it is characterized in that: in step (3), described rarefaction representation that all image blocks are carried out is that Large Scale Sparse represents, namely uses Batch-OMP Algorithm for Solving formula (1-1)
min||x-Dθ||2s.t.||θ||0≤T(1-1)
Wherein x is input signal, and D is that the single class obtained in step (2) represents dictionary or combines expression dictionary, and θ is coefficient, and T is openness condition.
5. the parasite egg recognition methods based on rarefaction representation according to claim 1, it is characterised in that: in step (4), described calculating reconstruction error matrix, step is as follows:
(1) utilize formula (1-2) to calculate reconstruction error, obtain the reconstruction error [e of all image blocks1,e2,…,eL], wherein L is image block number;
Wherein x is input signal, and D represents dictionary for single class or combines expression dictionary, and θ is coefficient;
(2) order of the image block obtained according to step (3), by [e1,e2,…,eL] arranged in sequence becomes a two-dimensional matrix, what obtain is reconstruction error matrix.
6. the parasite egg recognition methods based on rarefaction representation according to claim 1, it is characterised in that: in step (6), the sub-dictionary reconstruction error of described calculating, calculate according to formula (1-3)
Wherein x is input signal, DiFor associating dictionary D=[D1,D2,…,Dc] sub-dictionary, wherein i=1,2 ..., c, c is class number, and θ is coefficient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310181012.4A CN103268494B (en) | 2013-05-15 | 2013-05-15 | Parasite egg recognition methods based on rarefaction representation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310181012.4A CN103268494B (en) | 2013-05-15 | 2013-05-15 | Parasite egg recognition methods based on rarefaction representation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103268494A CN103268494A (en) | 2013-08-28 |
CN103268494B true CN103268494B (en) | 2016-06-15 |
Family
ID=49012121
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310181012.4A Expired - Fee Related CN103268494B (en) | 2013-05-15 | 2013-05-15 | Parasite egg recognition methods based on rarefaction representation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103268494B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015156844A1 (en) | 2014-04-10 | 2015-10-15 | MEP Equine Solutions LLC | Method for the quantification of parasite eggs in feces |
CN105138826A (en) * | 2015-08-10 | 2015-12-09 | 厦门大学 | Raman signal reconstruction method under strong noise background |
TWI703513B (en) * | 2019-01-31 | 2020-09-01 | 國立成功大學 | Egg counting device and method thereof |
CN110119776A (en) * | 2019-05-10 | 2019-08-13 | 长沙理工大学 | Recognition methods and its system based on Multiple Kernel Learning K-SVD |
CN111582276B (en) * | 2020-05-29 | 2023-09-29 | 北京语言大学 | Recognition method and system for parasite eggs based on multi-feature fusion |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102073872A (en) * | 2011-01-20 | 2011-05-25 | 中国疾病预防控制中心寄生虫病预防控制所 | Image-based method for identifying shape of parasite egg |
CN102891999A (en) * | 2012-09-26 | 2013-01-23 | 南昌大学 | Combined image compression/encryption method based on compressed sensing |
CN102968635A (en) * | 2012-11-23 | 2013-03-13 | 清华大学 | Image visual characteristic extraction method based on sparse coding |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006106508A2 (en) * | 2005-04-04 | 2006-10-12 | Technion Research & Development Foundation Ltd. | System and method for designing of dictionaries for sparse representation |
-
2013
- 2013-05-15 CN CN201310181012.4A patent/CN103268494B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102073872A (en) * | 2011-01-20 | 2011-05-25 | 中国疾病预防控制中心寄生虫病预防控制所 | Image-based method for identifying shape of parasite egg |
CN102891999A (en) * | 2012-09-26 | 2013-01-23 | 南昌大学 | Combined image compression/encryption method based on compressed sensing |
CN102968635A (en) * | 2012-11-23 | 2013-03-13 | 清华大学 | Image visual characteristic extraction method based on sparse coding |
Also Published As
Publication number | Publication date |
---|---|
CN103268494A (en) | 2013-08-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
CN109214399B (en) | Improved YOLOV3 target identification method embedded in SENET structure | |
TW201926140A (en) | Method, electronic device and non-transitory computer readable storage medium for image annotation | |
CN105608454B (en) | Character detecting method and system based on text structure component detection neural network | |
CN109447979B (en) | Target detection method based on deep learning and image processing algorithm | |
CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
CN106056595A (en) | Method for automatically identifying whether thyroid nodule is benign or malignant based on deep convolutional neural network | |
CN103268494B (en) | Parasite egg recognition methods based on rarefaction representation | |
CN102163281B (en) | Real-time human body detection method based on AdaBoost frame and colour of head | |
CN104850860A (en) | Cell image recognition method and cell image recognition device | |
CN108734138A (en) | A kind of melanoma skin disease image classification method based on integrated study | |
CN106778687A (en) | Method for viewing points detecting based on local evaluation and global optimization | |
CN110647875A (en) | Method for segmenting and identifying model structure of blood cells and blood cell identification method | |
CN103778435A (en) | Pedestrian fast detection method based on videos | |
CN110751232A (en) | Chinese complex scene text detection and identification method | |
CN108734200B (en) | Human target visual detection method and device based on BING (building information network) features | |
CN110969121A (en) | High-resolution radar target recognition algorithm based on deep learning | |
CN105512622A (en) | Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning | |
CN111582276A (en) | Parasite egg identification method and system based on multi-feature fusion | |
CN113096079B (en) | Image analysis system and construction method thereof | |
CN110969630A (en) | Ore bulk rate detection method based on RDU-net network model | |
CN109741351A (en) | A kind of classification responsive type edge detection method based on deep learning | |
CN117351371A (en) | Remote sensing image target detection method based on deep learning | |
CN103455798B (en) | Histogrammic human body detecting method is flowed to based on maximum geometry | |
CN105528791B (en) | A kind of quality evaluation device and its evaluation method towards touch screen hand-drawing image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160615 Termination date: 20170515 |