CN108921172B - Image processing device and method based on support vector machine - Google Patents
Image processing device and method based on support vector machine Download PDFInfo
- Publication number
- CN108921172B CN108921172B CN201810550987.2A CN201810550987A CN108921172B CN 108921172 B CN108921172 B CN 108921172B CN 201810550987 A CN201810550987 A CN 201810550987A CN 108921172 B CN108921172 B CN 108921172B
- Authority
- CN
- China
- Prior art keywords
- image
- image block
- level
- module
- support vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Abstract
The invention discloses an image processing device based on a support vector machine. The device includes: the image block acquisition module is used for acquiring an image block; the characteristic extraction module extracts the image block characteristics to obtain a characteristic vector, and comprises: a first extraction unit, a second extraction unit, a third extraction unit, and a fourth extraction unit; the image identification module is used for identifying whether the image block is normal or not according to whether a confidence score which is larger than a threshold value preset by a certain level exists in the certain level or not from a low level to a high level on the basis of a proximity algorithm; the image classification module classifies images based on a support vector machine; and the data processing module is used for processing the number of the identified image blocks and the number of the classified image blocks to finish the identification and classification of the image. The invention also discloses an image processing method based on the support vector machine. The invention can rapidly and accurately complete image recognition and classification, improves the accuracy of recognition and classification, and can be applied to pathological tissue image processing.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to an image processing apparatus and method based on a support vector machine.
Background
With the development of technologies such as multimedia, image resources are increasingly abundant, and compared with text resources, information contained in an image can often provide more data volume for a user, so that analyzing and managing the image becomes a hot spot of current research. At present, in the process of analyzing and managing images, the images are often required to be classified, that is, the images are divided into corresponding categories according to different characteristics reflected in image information.
At present, in the medical field, when images are identified and classified, pathological sections under microscopic fields of vision are identified, classified and evaluated mainly by experienced tissue pathologists, and the image identification and classification mode has various interferences under certain conditions, so that the image identification is inaccurate, and the accuracy of image classification is reduced. Therefore, it is a trend of today to increase the speed and accuracy of image processing.
Disclosure of Invention
In view of the above-mentioned drawbacks, an object of the present invention is to provide an image processing apparatus based on a support vector machine, which can improve the accuracy of image recognition and classification by performing image processing between modules cooperatively.
The invention also aims to provide an image processing method based on the support vector machine.
The above purpose is realized by the following technical scheme:
according to an aspect of the present invention, the present invention provides an image processing apparatus based on a support vector machine, including:
the image block acquisition module is used for acquiring an image block;
the characteristic extraction module is used for extracting the characteristics of the image block to obtain an image block characteristic vector; the feature extraction module includes: the first extraction unit is used for extracting local binary features of the image block during image identification; the second extraction unit is used for extracting the gray level co-occurrence matrix characteristics of the image blocks during image identification; the third extraction unit is used for extracting fractal features of the image blocks when the images are classified; the fourth extraction unit is used for extracting the directional gradient histogram characteristics of the image block when the image is classified;
the image recognition module is used for inputting image block feature vectors, performing image block recognition from a low level to a high level based on a proximity algorithm, and recognizing whether an image block is normal or not according to whether a confidence score greater than a threshold preset by a certain level exists in the certain level or not during recognition;
the image classification module classifies the image blocks according to the support vector machine;
a data processing module comprising: the first processing unit processes the number of the image blocks identified by the image identification module to finish the identification of the image, and the second processing unit processes the number of the image blocks classified by the image classification module to finish the classification of the image.
Preferably, the image block obtaining module includes: the first image block acquisition unit is used for acquiring image blocks during image identification, and the image blocks are subjected to hierarchical processing according to the resolution of the image blocks to obtain image blocks after hierarchical sampling; and the second image block acquisition unit acquires the image blocks during image classification in a local non-overlapping mode.
More preferably, the image block acquiring module includes a background image block marking unit, and marks the background image block according to a ratio of the total number of pixels to the number of pixels in which the three channel values of red, yellow, and blue of the image block are all greater than 200.
Preferably, the feature extraction module further includes a normalization processing unit, and the first extraction unit and the second extraction unit extract features, perform normalization processing, and then connect to obtain image block feature vectors used for image recognition. Wherein the normalization process is performed on the features [ f1, f2, … …, fN ] of the extracted image block]Where N represents the dimension of the extracted feature, the normalized values of each dimension
Wherein, miniIs the minimum value of the ith dimension, max, within the data setiIs the maximum value in the ith dimension in the data set.
Preferably, in the image recognition module, the threshold value preset by the lowest level is 0.9, and the threshold value is reduced by 0.1 per liter.
More preferably, when the image recognition module performs recognition, if a confidence score greater than a preset threshold of a level exists in the level, the image recognition module is an effective recognition image block and marks the category of the image block; if the image block cannot be identified at one level by a confidence score larger than a preset threshold value of the level, feature extraction and identification are carried out at the next level of the level, and if the image block does not have the next level, the label is not determined; if the image block cannot be identified with a confidence score greater than a threshold preset by the level at the highest level, the image block is an invalid identified image block.
Preferably, in the first processing unit, if the proportion of the number of valid recognized image blocks of a certain category in the total number of image blocks is greater than 50%, the picture is recognized as the category.
More preferably, the second processing unit determines whether the picture is classified into a certain class according to whether the number of image blocks in the certain class is judged to reach the corresponding level of the class and is the largest.
Preferably, the system further comprises a parameter selection module, wherein the parameter selection module is used for determining parameters in the image identification module and the image classification module.
According to another aspect of the present invention, the present invention provides a method for processing an image by using the above-mentioned image processing apparatus based on a support vector machine, including: image recognition and image classification, wherein,
the image recognition comprises the following steps: acquiring an image block, and carrying out hierarchical sampling on the image block; extracting local binary mode characteristics and gray level co-occurrence matrix characteristics under each level; performing normalization processing on the extracted features, and connecting to obtain image block feature vectors; for each image block feature vector, carrying out image identification based on a proximity algorithm from a low level to a high level;
the image classification includes the following steps: acquiring image blocks of the pictures in a non-overlapping mode; inputting an image block, and extracting fractal characteristics and directional gradient histogram characteristics; classifying based on a support vector machine, wherein a kernel function of the support vector machine is a radial basis function; and obtaining a picture classification result based on a voting mechanism.
Has the advantages that:
according to the invention, a first extraction unit and a second extraction unit in a feature extraction module perform local binary pattern feature (LBP feature) extraction and gray level co-occurrence matrix feature (Haralick feature) extraction on image blocks after hierarchical sampling, and the obtained feature vectors are subjected to image identification from low level to high level based on a proximity algorithm, namely K Nearest Neighbors (KNN), through an image identification module, so that the accuracy of image identification is improved; the third extraction unit and the fourth extraction unit perform fractal feature extraction and histogram of oriented gradient feature (HOG feature) extraction on the image blocks acquired in a non-overlapping manner, and then perform image classification based on a Support Vector Machine (SVM), so that the classification effect is good, and the accuracy of image classification is improved.
The image processing device and the method can quickly and accurately finish the identification and classification of the images, can provide a large amount of information such as related symptom types and related symptom grading information for pathologists when being applied to the processing of related histopathology images, and have great practical value in clinical medical treatment. The experimental result shows that the classification accuracy of the single image block can reach 95.27 percent; the accuracy rate of the normal class can reach 91.12%, the accuracy rate of the inflammation class is 68.07%, and the accuracy rate of the target symptom class is 56.59%.
Drawings
FIG. 1 is a schematic structural diagram of an image processing apparatus based on a support vector machine according to the present invention;
FIG. 2 is a schematic diagram of image recognition during image processing by the apparatus of the present invention;
FIG. 3 is a schematic diagram of image classification during image processing by the apparatus of the present invention;
FIG. 4 is a schematic diagram of the feature extraction process of the training set during image processing by the apparatus of the present invention;
fig. 5 is a schematic diagram of the preliminary effect of the device of the invention after image processing.
Detailed Description
The technical solution in the embodiment of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiment of the present invention:
wherein, fig. 1 schematically shows a structural diagram of an image processing apparatus based on a support vector machine according to the present invention; FIG. 2 schematically shows a process of image recognition using the image processing apparatus of the present invention; FIG. 3 schematically shows a process of image classification using the image processing apparatus of the present invention; FIG. 4 schematically shows a flow chart for performing training set sample feature extraction using the image processing apparatus of the present invention; fig. 5 schematically shows a preliminary effect diagram after image processing by the image processing apparatus of the present invention, wherein the lower right small diagram is used for displaying the thumbnail of the pathological tissue picture, and the large diagram is used for displaying the pathological tissue picture, including enlarging the specific part to see the details and conveniently marking the part with the pathological change.
As shown in fig. 1, the image processing apparatus based on support vector machine provided by the present invention includes: an image block obtaining module 100, configured to obtain image blocks, where each image block includes a training set image block and a sample set image block; the feature extraction module 200 is configured to extract features of an image block to obtain a feature vector of the image block; the feature extraction module 200 includes: the first extraction unit 210 is configured to, during image recognition, perform local binary feature extraction on the image block after the hierarchical sampling; the second extraction unit 220 is configured to, during image recognition, perform gray level co-occurrence matrix feature extraction on the image blocks after the hierarchical sampling; a third extraction unit 230, configured to perform fractal feature extraction on the image blocks obtained in a non-overlapping manner when classifying the images; a fourth extraction unit 240, configured to perform histogram feature extraction on gradient directions of the image blocks obtained in a non-overlapping manner when classifying the images; the image recognition module 300 is configured to input the obtained image block feature vector for image recognition, perform image block recognition from a low level to a high level based on a proximity algorithm, and recognize whether an image block is normal according to whether a confidence score greater than a threshold preset by a certain level exists in the certain level during recognition; the image classification module 400 classifies the obtained image block feature vectors for image classification based on a support vector machine; a data processing module 500 comprising: the first processing unit 510 performs processing on the number of image blocks recognized by the image recognition module 300 to complete the recognition of the image, and the second processing unit 520 performs processing on the number of classified image blocks of the image classification module 400 to complete the classification of the image.
In the device, after extracting local binary pattern features (LBP features) of a first extraction unit 210 and gray level co-occurrence matrix features (Haralick features) of a second extraction unit 220, the image is identified from a low level to a high level by a neighbor-based algorithm, namely a K Nearest Neighbors (KNN); after the fractal feature extraction of the third extraction unit 230 and the histogram of oriented gradient feature (HOG feature) extraction of the fourth unit 240, the image classification is performed based on a Support Vector Machine (SVM), and the method has the advantages of high image recognition and classification speed and high accuracy.
In an optional embodiment, the image block obtaining module 100 includes: the first image block acquisition unit is used for acquiring image blocks during image identification, and the image blocks are subjected to hierarchical processing according to the resolution of the image blocks to obtain image blocks after hierarchical sampling; a second image block acquisition unit which acquires an image block during image classification based on a non-overlapping manner; and the background image block marking unit marks according to the proportion of the total number of pixels to the number of pixels of which the red, yellow and blue three-channel values of the image block are all more than 200. Further, the first extraction unit 210 and the second extraction unit 220 extract features, perform normalization processing, and connect the features to obtain image block feature vectors used for image recognition.
In an alternative embodiment, the threshold preset for the lowest level in the image recognition module 300 is 0.9, and the threshold is decreased by 0.1 per liter. During identification, if a confidence score greater than a preset threshold of a level exists in the level, the image block is effectively identified, and the type of the image block is marked; if the image block cannot be identified at one level by a confidence score larger than a preset threshold value of the level, feature extraction and identification are carried out at the next level of the level, and if the image block does not have the next level, the label is not determined; if the image block cannot be identified with a confidence score greater than a threshold preset by the level at the highest level, the image block is an invalid identified image block.
In an alternative embodiment, in the data processing module 500, if the proportion of the number of valid recognized image blocks of a certain category in the total number of image blocks is greater than 50%, the picture is recognized as the category in the first processing unit. Whether the pictures are classified into a certain class is determined according to the judgment of whether the number of the image blocks in the certain class reaches the corresponding class of the class and is the largest.
In an alternative embodiment, the apparatus further comprises a parameter selection module for determining parameters in the image recognition module 300 and the image classification module 400. Through cross validation, the K value in the proximity algorithm is 27, the kernel function of the support vector machine is a radial basis function, the parameter C is 10, and the parameter gamma is 0.01.
As shown in fig. 2 to 4, the method for processing an image by using the image processing apparatus based on a support vector machine according to the present invention includes: image recognition and image classification, wherein the image recognition comprises the following steps: acquiring an image block, and carrying out hierarchical sampling on the image block; extracting local binary mode characteristics and gray level co-occurrence matrix characteristics under each level; performing normalization processing on the extracted features, and connecting to obtain image block feature vectors; for each image block feature vector, carrying out image identification based on a proximity algorithm from a low level to a high level; the image classification includes the following steps: acquiring image blocks of the pictures in a non-overlapping mode; inputting an image block, and extracting fractal characteristics and directional gradient histogram characteristics; classifying by adopting a support vector machine, wherein a kernel function of the support vector machine is a radial basis function; and obtaining a picture classification result based on a voting mechanism.
The following describes in detail the process of image processing by the image processing apparatus based on the support vector machine according to the present invention:
and (3) image identification process:
1) feature extraction for training set
The training samples are samples of known classes, and the non-parametric K nearest neighbor algorithm is adopted, so the model does not need to be trained, but training samples are needed as comparison standards, wherein fig. 4 is a flow chart of extracting the training samples as the following image block classification standard, and is also a process of performing image block classification and extracting each class of features, including performing multi-level image pyramid sampling on an image block, performing feature extraction through color space conversion to obtain a set [ f1, f2, … …, fN ], wherein N represents the dimension of the extracted features, and then performing feature selection to obtain a set [ f1, f2, … …, fsn ], wherein sn represents the dimension of the selected features after feature selection. The specific process is as follows:
each pathological tissue picture occupies a storage space which can reach several gigabits (Gb), firstly, training image blocks with the size of 896 × 896 pixels are obtained from the original pathological tissue picture, the step length is one-fourth of the width of the image block, wherein when the image blocks are extracted, if the number of pixels of which the RGB (red, yellow and blue) three-color channel values of one image block are all larger than 200 exceeds 70% of the total number of pixels, the image block is discarded as a background image block.
Then, LBP features and Haralick features are extracted. The image blocks need to be sampled in a grading manner before the characteristics are extracted, a multi-level image pyramid sampling mode is adopted, 3, 2, 1 and 0 times of down-sampling (the corresponding levels are 0, 1, 2 and 3) are sequentially carried out on the image blocks, the down-sampling actually is an image block operation, namely the size of the image is reduced by half according to nearest neighbor interpolation, bilinear interpolation or other interpolation modes, and the image is a python function when the actual implementation is realized. This is so at level 0, the function is called 3 times (or the "downsampling" operation is performed 3 times), at level 1, the function is called 2 times (or the "downsampling" operation is performed 2 times), and so on. At these levels, LBP features and Haralick features are extracted respectively through color space conversion, and the extraction of the features is realized by calling functions in an open-source algorithm library.
Then, respectively carrying out normalization processing on the LBP characteristic and the Haralick characteristic, wherein the normalization operation is carried out according to the following formula: for the features [ f1, f2, … …, fN ] of an extracted image block, N represents the dimension of the extracted feature, and the normalized values of the dimensions are as follows:
wherein, miniIs the minimum value of the ith dimension, max, within the data setiIs the maximum value of the ith dimension in the data set;
and performing normalization processing and connection, and finally expressing each image block in the training sample as a 4 multiplied by 1 multiplied by 73 dimensional matrix.
2) Feature extraction of new sample sets
The new sample is an unknown sample, the feature extraction method is the same as the feature extraction method of the training set, a plurality of image blocks with the same size are obtained for each pathological tissue picture with a few giga bits in a certain step length and half of the width of the image block, the background pixels also need to be marked in the process of obtaining the image block, if the proportion of the background pixels is more than 70 percent, the mark is that the background picture is not processed, and if the proportion of the background pixels is not more than 70 percent, the LBP feature and the Haralick feature are extracted from the lowest resolution. Thus, one pathological tissue picture of the new sample can be represented as a 4 × m × 73 dimensional matrix, where m represents the number of image blocks.
3) K value selection in K nearest neighbor algorithm
The K value was chosen to be 27 after 5-fold cross validation.
4) K nearest neighbor algorithm for identification marking
The idea of the K nearest neighbor algorithm is as follows: given a training set, for a new sample input instance, K instances (i.e., the most similar in feature space) that are closest to the new sample instance are found in the training set, and most of the K instances belong to a certain class, the new sample input instance is classified into the class. The K nearest neighbor algorithm of the invention has the following formula:
the specific identification and marking process of the K nearest neighbor algorithm in the invention comprises the following steps:
for each 4 × 1 × 73-dimensional vector in the new sample matrix, from the low level to the high level, the K-nearest neighbor algorithm is used to identify the image block, as shown in fig. 2.
I) if at a level, the K-nearest neighbor algorithm classifier marks the image block type as normal or abnormal (i.e. target symptom) if the confidence score is larger than the preset threshold of the level, and the image block is effectively identified.
Ii) if the image block can not be identified with the confidence score larger than the threshold value preset by the level at one level, performing feature extraction and identification at the next higher level of the level, if the next higher level does not exist, the label is not determined, and the image block is an invalid identification image block.
And iii) if the image block cannot be identified as a corresponding category with the confidence score larger than the threshold preset by the level at the highest level, the image block is an invalid identified image block, and finally the contribution of the image block is not considered when the category of the pathological tissue image is judged.
Wherein the preset threshold is a preset minimum value of the confidence score of each level.
In the present invention, the minimum confidence score for the initial level is set to 0.9, and one level per liter is higher, the minimum confidence score is reduced by 0.1. That is, at level 0, the confidence score minimum (or threshold) needs to be 0.9 and level 1 minimum needs to be 0.8, with each subsequent increase in level the confidence score minimum decreasing by 0.1. For example, in step i), the image block class is labeled only when the level 0 needs to satisfy the confidence score greater than 0.9 and the level 1 needs to satisfy the confidence score greater than 0.8, otherwise, the image block class is identified according to steps ii) and iii).
5) Judgment of recognition result
If the number of valid judgment image blocks in a certain category is m _ correct, the total number of image blocks is m _ all, and the ratio thereof
More than 50%, the pathological tissue picture is identified as the category。
The image classification process is shown in fig. 3:
in the invention, the image classification feature extraction and the image recognition feature extraction are different, and the specific classification process is as follows: the image classification is known when the training set is extracted, and is unknown when the image classification is carried out after the classification model is trained.
11) Feature extraction for training set
Firstly, acquiring 384 multiplied by 384 image blocks in a non-overlapping mode aiming at a large pathological tissue picture, wherein in the image block acquisition process, a background image block needs to be discarded, the background image block judgment mode is the same as the mode of a k-nearest neighbor algorithm, and if the proportion of background pixels is more than 70%, the background image blocks are marked as background pictures and are not processed.
And then, extracting fractal features and extracting directional gradient histogram descriptor features of the obtained image blocks, wherein the extraction of the directional gradient histogram descriptor features and the extraction of the fractal features both adopt functions in an open source algorithm library, and each extracted image block is expressed as a vector of 1 × 44 dimensions.
12) Pathological tissue picture feature extraction of unknown classes
For one pathological tissue picture, the feature extraction method is similar to the feature extraction method in step 11), and after image blocks are obtained and features are extracted, each pathological tissue picture is expressed as an m × 44 dimensional matrix, wherein m is the number of the image blocks.
13) Hyper-parameter selection
The image classification is carried out by adopting a support vector machine, the kernel function of the support vector machine is selected as a radial basis function, so that a better C value and a better gamma are required to be selected, the parameter C is 10 and the parameter gamma is 0.01 after 3-fold cross validation, wherein the parameter C and the gamma are parameters of the support vector machine, and the parameters are set as fixed values after the support vector machine is trained.
14) Image classification
And classifying by adopting a support vector machine, and obtaining the image category based on a voting mechanism.
If the number of the judging image blocks of one type in the judged types reaches m _ class and is the maximum, each type has m _ class which represents the number of the image blocks classified into the type, the pathological tissue picture is judged as the type, and the pathological tissue picture in the step 12) is marked.
The method carries out image identification based on K nearest neighbor algorithm on the basis of local binary mode feature extraction and gray level co-occurrence matrix feature extraction, then carries out image classification based on support vector machine on the basis of fractal feature extraction and directional gradient histogram feature extraction, can rapidly and accurately complete image identification and classification, and has the classification accuracy rate of a single image block up to 95.27%; the invention is applied to the treatment of pathological tissue pictures, can provide a large amount of information for histopathologists, the experimental result shows, when classifying each pathological tissue picture, the classification accuracy rate of the normal image class is 91.12%, the classification accuracy rate of the inflammation image class is 68.07%, and the classification accuracy rate of the target symptom image class is 56.69%.
While the preferred embodiments of the present invention have been illustrated and described, it will be appreciated by those skilled in the art that the foregoing embodiments are illustrative and not restrictive, and that many changes may be made in the embodiment without departing from the spirit and the scope of the appended claims.
Claims (9)
1. An image processing apparatus based on a support vector machine, comprising:
the image block acquisition module is used for acquiring an image block;
the characteristic extraction module is used for extracting the characteristics of the image block to obtain an image block characteristic vector; the feature extraction module includes: the first extraction unit is used for extracting local binary features of the image block during image identification; the second extraction unit is used for extracting the gray level co-occurrence matrix characteristics of the image blocks during image identification; the third extraction unit is used for extracting fractal features of the image blocks when the images are classified; the fourth extraction unit is used for extracting the directional gradient histogram characteristics of the image block when the image is classified;
the image recognition module is used for inputting image block feature vectors, carrying out image block recognition from a low level to a high level according to the resolution of an image block based on a proximity algorithm, and recognizing whether the image block is normal or not according to whether a confidence score greater than a threshold preset by a certain level exists in the certain level or not during recognition;
the image classification module is used for classifying the image blocks based on a support vector machine;
a data processing module comprising: the first processing unit processes the number of the image blocks identified by the image identification module to finish the identification of the image, and the second processing unit processes the number of the image blocks classified by the image classification module to finish the classification of the image.
2. The support vector machine-based image processing apparatus of claim 1, wherein the image block obtaining module comprises: the first image block acquisition unit is used for acquiring image blocks during image identification, and the image blocks are subjected to hierarchical processing according to the resolution of the image blocks to obtain image blocks after hierarchical sampling; and the second image block acquisition unit acquires the image blocks during image classification based on a non-overlapping mode.
3. The SVM based image processing device of claim 2, wherein the image block acquiring module includes a background image block labeling unit labeling the image block according to a ratio of the total number of pixels to the number of pixels in which three channel values of red, yellow and blue of the image block are all greater than 200.
4. The support vector machine-based image processing apparatus of claim 1, wherein the feature extraction module further comprises a normalization processing unit, the first extraction unit and the second extraction unit perform normalization processing after extracting features and then connect to obtain feature vectors of image blocks, wherein the normalization processing is performed on the extracted features [ f1, f2, … …, fN ] of an image block, where N represents the dimension of the extracted features, and the normalized dimensions have values of:
wherein, miniIs the minimum value of the ith dimension, max, within the data setiIs the maximum value in the ith dimension within the data set.
5. The support vector machine-based image processing apparatus according to claim 1, wherein in the image recognition module, the threshold value preset by the lowest level is 0.9, and the threshold value is decreased by 0.1 per liter.
6. The SVM based image processing device of claim 5, wherein the image recognition module, when recognizing, if there is a confidence score greater than a preset threshold at a level, marks the image block type for effective recognition of the image block; if the image block cannot be identified at one level by a confidence score larger than a preset threshold value of the level, feature extraction and identification are carried out at the next level of the level, and if the image block does not have the next level, the label is not determined; if the image block cannot be identified with a confidence score greater than a threshold preset by the level at the highest level, the image block is an invalid identified image block.
7. The support vector machine-based image processing apparatus of claim 1, wherein in the first processing unit, if the number of valid recognized image blocks of a certain category accounts for more than 50% of the total number of image blocks, the picture is recognized as the category.
8. The support vector machine-based image processing apparatus of claim 7, wherein the second processing unit determines whether the picture is classified into a class according to whether the number of image blocks in the class is the largest and reaches the corresponding class of the class.
9. The support vector machine-based image processing apparatus of claim 1, wherein the image processing apparatus further comprises a parameter selection module for determining parameters in the image recognition module and the image classification module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810550987.2A CN108921172B (en) | 2018-05-31 | 2018-05-31 | Image processing device and method based on support vector machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810550987.2A CN108921172B (en) | 2018-05-31 | 2018-05-31 | Image processing device and method based on support vector machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108921172A CN108921172A (en) | 2018-11-30 |
| CN108921172B true CN108921172B (en) | 2021-01-12 |
Family
ID=64410687
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810550987.2A Active CN108921172B (en) | 2018-05-31 | 2018-05-31 | Image processing device and method based on support vector machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921172B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109949269A (en) * | 2019-01-24 | 2019-06-28 | 武汉华目信息技术有限责任公司 | A kind of detection method and device of railroad train dust cap breakage failure |
| CN110223303A (en) * | 2019-05-13 | 2019-09-10 | 清华大学 | HE dyes organ pathological image dividing method, device |
| CN113132704B (en) * | 2020-01-15 | 2022-08-19 | 华为技术有限公司 | Image processing method, device, terminal and storage medium |
| CN113237818A (en) * | 2021-05-28 | 2021-08-10 | 上海睿钰生物科技有限公司 | Cell analysis method and system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103489178A (en) * | 2013-08-12 | 2014-01-01 | 中国科学院电子学研究所 | Method and system for image registration |
| US8885901B1 (en) * | 2013-10-22 | 2014-11-11 | Eyenuk, Inc. | Systems and methods for automated enhancement of retinal images |
| CN106503728A (en) * | 2016-09-30 | 2017-03-15 | 深圳云天励飞技术有限公司 | A kind of image-recognizing method and device |
-
2018
- 2018-05-31 CN CN201810550987.2A patent/CN108921172B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN108921172A (en) | 2018-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3455782B1 (en) | System and method for detecting plant diseases | |
| CN108921172B (en) | Image processing device and method based on support vector machine | |
| CN106803247B (en) | Microangioma image identification method based on multistage screening convolutional neural network | |
| US8379961B2 (en) | Mitotic figure detector and counter system and method for detecting and counting mitotic figures | |
| US10817741B2 (en) | Word segmentation system, method and device | |
| CN108090906B (en) | Cervical image processing method and device based on region nomination | |
| CN109784342B (en) | OCR (optical character recognition) method and terminal based on deep learning model | |
| EP3203417B1 (en) | Method for detecting texts included in an image and apparatus using the same | |
| US20110085728A1 (en) | Detecting near duplicate images | |
| CN111145209A (en) | Medical image segmentation method, device, equipment and storage medium | |
| CN108009567B (en) | Automatic excrement character distinguishing method combining image color and HOG and SVM | |
| WO2019204577A1 (en) | System and method for multimedia analytic processing and display | |
| CN110751619A (en) | Insulator defect detection method | |
| CN112464983A (en) | Small sample learning method for apple tree leaf disease image classification | |
| CN111126401A (en) | License plate character recognition method based on context information | |
| Xue et al. | Curved text detection in blurred/non-blurred video/scene images | |
| CN112703531A (en) | Generating annotation data for tissue images | |
| CN112365451A (en) | Method, device and equipment for determining image quality grade and computer readable medium | |
| Rani et al. | Object Detection in Natural Scene Images Using Thresholding Techniques | |
| Mia et al. | CompTRNet: An U-Net Approach for White Blood Cell Segmentation Using Compound Loss Function and Transfer Learning with Pre-trained ResNet34 Network | |
| CN113255665B (en) | Target text extraction method and system | |
| CN111832626B (en) | Image recognition classification method, device and computer readable storage medium | |
| Monteiro et al. | Deep learning recognition of a large number of pollen grain types | |
| Murthy et al. | A Novel method for efficient text extraction from real time images with diversified background using haar discrete wavelet transform and k-means clustering | |
| CN113076438B (en) | Classification method based on conversion from majority class to minority class under unbalanced data set |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |