CN109993201A - A kind of image processing method, device and readable storage medium storing program for executing - Google Patents
A kind of image processing method, device and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN109993201A CN109993201A CN201910114848.XA CN201910114848A CN109993201A CN 109993201 A CN109993201 A CN 109993201A CN 201910114848 A CN201910114848 A CN 201910114848A CN 109993201 A CN109993201 A CN 109993201A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- classifier
- neural networks
- convolutional neural
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 56
- 230000002708 enhancing effect Effects 0.000 claims abstract description 24
- 238000005516 engineering process Methods 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims abstract description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 96
- 238000013480 data collection Methods 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 20
- 238000000034 method Methods 0.000 abstract description 27
- 230000001965 increasing effect Effects 0.000 abstract description 9
- 230000008569 process Effects 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 239000013598 vector Substances 0.000 description 28
- 238000012706 support-vector machine Methods 0.000 description 17
- 230000006870 function Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 230000001537 neural effect Effects 0.000 description 4
- 241000894007 species Species 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 2
- 241000282326 Felis catus Species 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to artificial intelligence technology, a kind of image processing method, device and readable storage medium storing program for executing are specifically disclosed, wherein method includes: acquisition image information;According to described image information, feature extraction is carried out using feature extractor, obtains characteristic information;The characteristic information is sent to classifier;The classifier is classified according to the characteristic information, obtains classification information.According to the technical solution of the present invention, the minimal characteristic distance in classifier is judged, when being greater than preset distance threshold just using classification information as final classification results, the accuracy of image classification can be increased.And the present invention carries out the extraction of characteristic information using the convolutional layer in neural network as feature extractor, can further increase the accuracy of identification.The present invention carries out modal distortion enhancing also in image training process, for image, further increases the accuracy of image recognition.
Description
Technical field
The present invention relates to the technical field of image processing of artificial intelligence field, in particular to one kind based on depth mind
Image processing method, a kind of terminal and a kind of computer volume readable storage medium storing program for executing through network.
Background technique
With the high speed development of machine learning techniques, convolutional neural networks are increasingly used in computer vision,
Especially image classification field.
Convolutional neural networks (CNN) are a kind of powerful machine learning techniques used in the deep learning field.Simply
Neural network be used as the machine learning techniques of character recognition and image classification field.However, such a nerve of training
Network needs a large amount of flag data.In the case where not spending time and efforts to be trained, the simple side of CNN function is utilized
Method is to use CNN trained in advance as feature extractor.In order to generate complicated decision curved surface, support vector machines (SVM) is
A kind of very economical method, for indicating complex-curved in higher dimensional space, including multinomial and other kinds of curved surface.
In the related technology, convolutional neural networks are a kind of powerful machine learning skills used in the deep learning field
Art.Simple convolutional neural networks have been used as the machine learning techniques of character recognition and image classification field.But existing skill
It is not high to the nicety of grading of image using simple CNN and SVM technology in art, so designing a kind of skill that can combine the two
Art be used to promoted image recognition nicety of grading scheme be urgently can not to.
Summary of the invention
In order to solve at least one above-mentioned technical problem, the invention proposes a kind of image processing methods, device and readable
Storage medium.
First aspect present invention provides a kind of image processing method, comprising:
Obtain image information;
According to described image information, feature extraction is carried out using feature extractor, obtains characteristic information;
The characteristic information is sent to classifier;
Classifier is calculated according to the characteristic information, obtains characteristic information and hyperplane in corresponding classifier most
Small characteristic distance;
Judge whether the minimal characteristic distance is greater than preset distance threshold;
If more than then using the classification information of the classifier as final classification information.
In the present solution, the feature extractor is the feature extractor by training setting in advance, specifically:
Obtain extensive object image data collection;
The extensive object image data collection is subjected to image training, obtains depth convolutional neural networks AlexNet;
Obtain the convolutional layer of the depth convolutional neural networks AlexNet;
Using the convolutional layer as feature extractor.
In the present solution, the first layer of depth convolutional neural networks AlexNet includes the filter for capturing edge feature,
The middle layer of the depth convolutional neural networks AlexNet includes multiple convolutional layers and maximum pond layer, the depth convolutional Neural
The last layer of network A lexNet is classification layer, and the weight of each layer of the depth convolutional neural networks AlexNet passes through training
It determines.
It, will be in the present solution, be used as feature extractor using the 17th layer of the depth convolutional neural networks AlexNet
The 17th layer of obtained characteristic information of the depth convolutional neural networks AlexNet is sent in classifier.
In the present solution, the classifier is the classifier by training setting in advance.
In the present solution, described carry out image training for the extensive object image data collection, depth convolutional Neural is obtained
Network A lexNet, further includes:
It is handled using the image concentrated to extensive object image data of mode enhancing technology, obtains that treated schemes
Picture;
Treated that image is trained for described, obtains new depth convolutional neural networks AlexNet.
Specifically, the mode enhancing technology is specially one of rotation, inclination, elastic distortion, cosine function enhancing
Or it is several.
Second aspect of the present invention provides a kind of image processing apparatus, which includes: memory, processor, the storage
Include image processing method program in device, described image processing method program realizes following steps when being executed by the processor:
Obtain image information;
According to described image information, feature extraction is carried out using feature extractor, obtains characteristic information;
The characteristic information is sent to classifier;
Classifier is calculated according to the characteristic information, obtains characteristic information and hyperplane in corresponding classifier most
Small characteristic distance;
Judge whether the minimal characteristic distance is greater than preset distance threshold;
If more than then using the classification information of the classifier as final classification information.
In the present solution, the feature extractor is the feature extractor by training setting in advance, specifically:
Obtain extensive object image data collection;
The extensive object image data collection is subjected to image training, obtains depth convolutional neural networks AlexNet;
Obtain the convolutional layer of the depth convolutional neural networks AlexNet;
Using the convolutional layer as feature extractor.
In the present solution, the first layer of depth convolutional neural networks AlexNet includes the filter for capturing edge feature,
The middle layer of the depth convolutional neural networks AlexNet includes multiple convolutional layers and maximum pond layer, the depth convolutional Neural
The last layer of network A lexNet is classification layer, and the weight of each layer of the depth convolutional neural networks AlexNet passes through training
It determines.
It, will be in the present solution, be used as feature extractor using the 17th layer of the depth convolutional neural networks AlexNet
The 17th layer of obtained characteristic information of the depth convolutional neural networks AlexNet is sent in classifier.
In the present solution, the classifier is the classifier by training setting in advance.
In the present solution, described carry out image training for the extensive object image data collection, depth convolutional Neural is obtained
Network A lexNet, further includes:
It is handled using the image concentrated to extensive object image data of mode enhancing technology, obtains that treated schemes
Picture;
Treated that image is trained for described, obtains new depth convolutional neural networks AlexNet.
Specifically, the mode enhancing technology is specially one of rotation, inclination, elastic distortion, cosine function enhancing
Or it is several.
Third aspect present invention provides a kind of computer readable storage medium, includes in the computer readable storage medium
Image processing method program when described image processing method program is executed by processor, realizes a kind of such as above-mentioned image procossing
The step of method.
Image processing method, device and the readable storage medium storing program for executing being related to through the invention, to the minimal characteristic in classifier
Distance is judged, when being greater than preset distance threshold just using classification information as final classification results, can increase figure
As the accuracy of classification.And the convolutional layer in neural network is carried out mentioning for characteristic information by the present invention
It takes, the accuracy of identification can be further increased.The present invention carries out modal distortion increasing also in image training process, for image
By force, the accuracy of image recognition is further increased.
Additional aspect and advantage of the invention will provide in following description section, will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect of the invention and advantage will become from the description of the embodiment in conjunction with the following figures
Obviously and it is readily appreciated that, in which:
Fig. 1 shows a kind of flow chart of image processing method of the present invention;
Fig. 2 shows the layer structure charts of depth convolutional neural networks AlexNet of the present invention;
Fig. 3 shows the first layer weighting structure figure of depth convolutional neural networks AlexNet of the present invention;
Fig. 4 shows the schematic diagram that the present invention carries out image enhancement by cosine mode;
Fig. 5 shows the schematic diagram that the present invention carries out image enhancement by rotation inclination;
Fig. 6 shows a kind of structural block diagram of image processing apparatus of the present invention.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
To be implemented using other than the one described here other modes, therefore, protection scope of the present invention is not by described below
Specific embodiment limitation.
Fig. 1 shows a kind of flow chart of image processing method of the present invention.
As shown in Figure 1, the invention discloses a kind of image processing methods, comprising:
S102 obtains image information;
S104 carries out feature extraction using feature extractor, obtains characteristic information according to described image information;
The characteristic information is sent to classifier by S106;
S108, classifier are calculated according to the characteristic information, and it is super flat with corresponding classifier to obtain characteristic information
The minimal characteristic distance in face;
S110, judges whether the minimal characteristic distance is greater than preset distance threshold;
S112, if more than then using the classification information of the classifier as final classification information.
It should be noted that after obtaining picture, that is, obtain image information, just input feature vector extractor into
Row feature extraction, obtains characteristic information.That is, the original image of input is transformed to feature vector by feature extractor, institute
Stating feature vector is characteristic information.Then characteristic information is sent in classifier, classification processing is carried out, finally by input
Image output is multiple categorised contents.For example, one picture comprising kitten of input, then first obtain the image information of this picture,
Then the extraction that feature is carried out according to this image information, obtains characteristic information, characteristic information is sent in classifier and is divided
Class identification, final output classification information shows that the recognition result of picture for kitten, and can mark kitten institute in classification information
Position in picture.Certainly, classification information, for example, marking detection block in picture, can also be detected comprising other information
The position of frame delineation species;Species information etc. is marked in detection block.Those skilled in the art can set according to actual needs
The type of classification information, but any method based on technical solution of the present invention output category information falls within protection of the present invention
In range.
It should be noted that classifier, such as SVM classifier, it is the identification and classification device defined by Optimal Separating Hyperplane.
That is the training sample of one group of tape label is given, algorithm will export an optimal hyperlane to new samples (test sample)
Classify.That is, trained classifier, there are a hyperplane, this hyperplane divides each feature vector
It is divided into different classifications, the feature vector in each region is better further away from hyperplane, enables to classification more accurate in this way.
When feature vector falls into some region, whether this minimum range can be judged with the hyperplane there are a minimum range
Greater than preset distance threshold, if being less than preset threshold value, shows there may be certain error, image can be carried out and located in advance
Then reason re-starts feature vector and calculates and classify;If more than then showing that classification results are accurate, classification information can be made
For final classification results.Wherein pretreatment may include the operations such as contrast increases, color increases, eliminates picture noise.
According to embodiments of the present invention, the feature extractor is the feature extractor by training setting in advance, specifically:
Obtain extensive object image data collection;
The extensive object image data collection is subjected to image training, obtains depth convolutional neural networks AlexNet;
Obtain the convolutional layer of the depth convolutional neural networks AlexNet;
Using the convolutional layer as feature extractor.
It should be noted that said features extractor is pre-set feature extractor.The feature extractor is pre-
The first convolutional layer of trained depth convolutional neural networks AlexNet.It is specific in training depth convolutional neural networks AlexNet
Are as follows: extensive object image data collection is obtained, which has 1000 object type and 1,200,000 instructions
Practice image.Image training is carried out for the extensive object image data collection, obtains depth convolutional neural networks AlexNet;
Then the convolutional layer information of the depth convolutional neural networks AlexNet is obtained, wherein depth convolutional neural networks AlexNet has
Standby multiple convolutional layers, the present invention only needs to obtain wherein one layer of convolutional layer, as feature extractor.
Specifically, Fig. 2 shows the layer structure charts of depth convolutional neural networks AlexNet of the present invention.As shown in Fig. 2, deep
Degree convolutional neural networks AlexNet input is an image in 1000 different type images (such as cat, dog), and output is
1000 digital vectors.I-th of element of output vector is the probability that input picture belongs to the i-th class image.Wherein, depth
The size of input picture is defined as 227 × 227 × 3 by the first layer of convolutional neural networks AlexNet.Layer is for extensive again
The training in advance of object image data collection.The first layer of network learns the filter for capturing blob and edge feature.It is intermediate
Layer is a series of five convolutional layers and three layers being fully connected, and is scattered with rectification linear unit (ReLU) and maximum pond layer.Most
Later layer is classification layer, there is 1000 classes.Pre-training CNN for object image data collection is used as feature extractor.Depth volume
The first convolutional layer weight of product neural network AlexNet is as shown in Figure 3.It is not the object diagram in training such as STL-10 database
Picture, but the CNN of object images pre-training (i.e. Alex-Net) is used as feature extractor.Depth convolutional neural networks
The 17th layer of entitled " fc7 " is connected to classifier as feature extractor in AlexNet layers of figure, that is, will obtain at the 17th layer
Feature vector be passed to classifier.
According to embodiments of the present invention, the first layer of depth convolutional neural networks AlexNet includes for capturing edge feature
Filter, the middle layer of the depth convolutional neural networks AlexNet includes multiple convolutional layers and maximum pond layer, the depth
The last layer of convolutional neural networks AlexNet is classification layer, the weight of each layer of the depth convolutional neural networks AlexNet
It is determined by training.
According to embodiments of the present invention, it is mentioned using the 17th layer of the depth convolutional neural networks AlexNet as feature
Device is taken, will be sent in classifier in the 17th layer of obtained characteristic information of the depth convolutional neural networks AlexNet.
It should be noted that the present invention is used as feature using the 17th layer of the depth convolutional neural networks AlexNet
Extractor, wherein being specially the layer for naming " fc7 " in depth convolutional neural networks AlexNet, that is, in AlexNet network
The layer of " fc7 " in the full articulamentum of fully connected.The feature vector obtained at the 17th layer is passed to classifier
In.
According to embodiments of the present invention, the classifier is the classifier by training setting in advance.
It should be noted that classifier can be carried out using support vector machines (Support Vector Machine, SVM)
Classification.The classifier comes by searching for the best hyperplane for separating the data point of all data points of class and other classes
Classify to data, support vector machines (classifier) can indicate complicated surface, including multinomial and radial basis function.Most
Good hyperplane is maximum hyperplane between two classes, and in other words, the maximum that back gauge is parallel to the plate of hyperplane is wide
Degree, does not have internal data point, wherein supporting vector is closest to the data point of separating hyperplance.
It should be noted that using convolutional neural networks functional training classifier.The image feature information that will acquire, also
It is that feature vector input classifier is trained.When using higher-dimension CNN feature vector (each length be 4096 layers), using with
Machine gradient declines (SGD) solver to accelerate to train.In order to measure the classification accuracy of trained SVM classifier, mentioned by CNN
It takes test image feature and passes it to SVM classifier.It, can be with using the classifier after training as the classifier of image recognition
Improve the accuracy rate of image recognition.
According to embodiments of the present invention, described that the extensive object image data collection is subjected to image training, obtain depth
Convolutional neural networks AlexNet, further includes:
It is handled using the image concentrated to extensive object image data of mode enhancing technology, obtains that treated schemes
Picture;
Treated that image is trained for described, obtains new depth convolutional neural networks AlexNet.
It should be noted that when obtaining depth convolutional neural networks AlexNet, can also will scheme carrying out image training
As carrying out enhancing processing.Enhancing processing mentioned here, can be modal distortion enhancing.Data enhancement methods are wherein used as, are imitated
It penetrates transformation and elasticity distortion is well-known in character recognition.They are used to generate new samples from original sample and extend instruction
Practice collection.By the way that affine displacement field is applied to mode, it can be generated and such as translate, rotate, scale and inclined simple distortion.Bullet
Property distortion be it is a kind of imitation handwriting style variation image transformation.
Specifically, the mode enhancing technology is specially one of rotation, inclination, elastic distortion, cosine function enhancing
Or it is several.It should be understood by those skilled in the art that the present invention not merely limit more than enhancement mode, it is any through the invention
Technical solution carry out mode enhancing method fall in the scope of the present invention.
Fig. 4 shows the schematic diagram that the present invention carries out image enhancement by cosine mode.
As shown in figure 4, by the present invention in that with the mode conversion method with cosine function.Specifically also apply some increasings
Strong method such as rotates, inclination and elastic distortion.Fig. 4 shows the mode example of cosine function enhancing.The quantity of training image increases
31 times of original 5k training sample are added.By using cosine function, raw mode is left-justify, Right Aligns, top alignment or bottom
Alignment, mode be also center alignment and it is widened.
Fig. 5 shows the schematic diagram that the present invention carries out image enhancement by rotation inclination.
As shown in figure 5, image, is carried out the rotation of certain angle by rotation processing of the present invention by using image;Or
Using the inclined processing of image, purpose is to be transformed to image the image of certain angle is presented with original image.It can also lead to
The mode for crossing elastic distortion carries out image procossing, and original image is transformed to the image of distortion.
By using treated, image is trained, and obtains new depth convolutional neural networks AlexNet.The present invention is logical
Crossing treated, certain distorted image is trained, and can increase the probability of matching image in a network, is reached and is improved image and know
The purpose of other accuracy.
Fig. 6 shows a kind of structural block diagram of image processing apparatus of the present invention.
As shown in fig. 6, second aspect of the present invention provides a kind of image processing apparatus, which includes: memory 61, processing
Device 62 includes image processing method program in the memory, when described image processing method program is executed by the processor
Realize following steps:
Obtain image information;
According to described image information, feature extraction is carried out using feature extractor, obtains characteristic information;
The characteristic information is sent to classifier;
Classifier is calculated according to the characteristic information, obtains characteristic information and hyperplane in corresponding classifier most
Small characteristic distance;
Judge whether the minimal characteristic distance is greater than preset distance threshold;
If more than then using the classification information of the classifier as final classification information.
It should be noted that after obtaining picture, that is, obtain image information, just input feature vector extractor into
Row feature extraction, obtains characteristic information.That is, the original image of input is transformed to feature vector by feature extractor, institute
Stating feature vector is characteristic information.Then characteristic information is sent in classifier, classification processing is carried out, finally by input
Image output is multiple categorised contents.For example, one picture comprising kitten of input, then first obtain the image information of this picture,
Then the extraction that feature is carried out according to this image information, obtains characteristic information, characteristic information is sent in classifier and is divided
Class identification, final output classification information shows that the recognition result of picture for kitten, and can mark kitten institute in classification information
Position in picture.Certainly, classification information, for example, marking detection block in picture, can also be detected comprising other information
The position of frame delineation species;Species information etc. is marked in detection block.Those skilled in the art can set according to actual needs
The type of classification information, but any method based on technical solution of the present invention output category information falls within protection of the present invention
In range.
It should be noted that classifier, such as SVM classifier, it is the identification and classification device defined by Optimal Separating Hyperplane.
That is the training sample of one group of tape label is given, algorithm will export an optimal hyperlane to new samples (test sample)
Classify.That is, trained classifier, there are a hyperplane, this hyperplane divides each feature vector
It is divided into different classifications, the feature vector in each region is better further away from hyperplane, enables to classification more accurate in this way.
When feature vector falls into some region, whether this minimum range can be judged with the hyperplane there are a minimum range
Greater than preset distance threshold, if being less than preset threshold value, shows there may be certain error, image can be carried out and located in advance
Then reason re-starts feature vector and calculates and classify;If more than then showing that classification results are accurate, classification information can be made
For final classification results.Wherein pretreatment may include the operations such as contrast increases, color increases, eliminates picture noise.
According to embodiments of the present invention, the feature extractor is the feature extractor by training setting in advance, specifically:
Obtain extensive object image data collection;
The extensive object image data collection is subjected to image training, obtains depth convolutional neural networks AlexNet;
Obtain the convolutional layer of the depth convolutional neural networks AlexNet;
Using the convolutional layer as feature extractor.
It should be noted that said features extractor is pre-set feature extractor.The feature extractor is pre-
The first convolutional layer of trained depth convolutional neural networks AlexNet.It is specific in training depth convolutional neural networks AlexNet
Are as follows: extensive object image data collection is obtained, which has 1000 object type and 1,200,000 instructions
Practice image.Image training is carried out for the extensive object image data collection, obtains depth convolutional neural networks AlexNet;
Then the convolutional layer information of the depth convolutional neural networks AlexNet is obtained, wherein depth convolutional neural networks AlexNet has
Standby multiple convolutional layers, the present invention only needs to obtain wherein one layer of convolutional layer, as feature extractor.
Specifically, Fig. 2 shows the layer structure charts of depth convolutional neural networks AlexNet of the present invention.As shown in Fig. 2, deep
Degree convolutional neural networks AlexNet input is an image in 1000 different type images (such as cat, dog), and output is
1000 digital vectors.I-th of element of output vector is the probability that input picture belongs to the i-th class image.Wherein, depth
The size of input picture is defined as 227 × 227 × 3 by the first layer of convolutional neural networks AlexNet.Layer is for extensive again
The training in advance of object image data collection.The first layer of network learns the filter for capturing blob and edge feature.It is intermediate
Layer is a series of five convolutional layers and three layers being fully connected, and is scattered with rectification linear unit (ReLU) and maximum pond layer.Most
Later layer is classification layer, there is 1000 classes.Pre-training CNN for object image data collection is used as feature extractor.Depth volume
The first convolutional layer weight of product neural network AlexNet is as shown in Figure 3.It is not the object diagram in training such as STL-10 database
Picture, but the CNN of object images pre-training (i.e. Alex-Net) is used as feature extractor.Depth convolutional neural networks
The 17th layer of entitled " fc7 " is connected to classifier as feature extractor in AlexNet layers of figure, that is, will obtain at the 17th layer
Feature vector be passed to classifier.
According to embodiments of the present invention, the first layer of depth convolutional neural networks AlexNet includes for capturing edge feature
Filter, the middle layer of the depth convolutional neural networks AlexNet includes multiple convolutional layers and maximum pond layer, the depth
The last layer of convolutional neural networks AlexNet is classification layer, the weight of each layer of the depth convolutional neural networks AlexNet
It is determined by training.
According to embodiments of the present invention, it is mentioned using the 17th layer of the depth convolutional neural networks AlexNet as feature
Device is taken, will be sent in classifier in the 17th layer of obtained characteristic information of the depth convolutional neural networks AlexNet.
It should be noted that the present invention is used as feature using the 17th layer of the depth convolutional neural networks AlexNet
Extractor, wherein being specially the layer for naming " fc7 " in depth convolutional neural networks AlexNet, that is, in AlexNet network
The layer of " fc7 " in the full articulamentum of fully connected.The feature vector obtained at the 17th layer is passed to classifier
In.
According to embodiments of the present invention, the classifier is the classifier by training setting in advance.
It should be noted that classifier can be carried out using support vector machines (Support Vector Machine, SVM)
Classification.The classifier comes by searching for the best hyperplane for separating the data point of all data points of class and other classes
Classify to data, support vector machines (classifier) can indicate complicated surface, including multinomial and radial basis function.Most
Good hyperplane is maximum hyperplane between two classes, and in other words, the maximum that back gauge is parallel to the plate of hyperplane is wide
Degree, does not have internal data point, wherein supporting vector is closest to the data point of separating hyperplance.
It should be noted that using convolutional neural networks functional training classifier.The image feature information that will acquire, also
It is that feature vector input classifier is trained.When using higher-dimension CNN feature vector (each length be 4096 layers), using with
Machine gradient declines (SGD) solver to accelerate to train.In order to measure the classification accuracy of trained SVM classifier, mentioned by CNN
It takes test image feature and passes it to SVM classifier.It, can be with using the classifier after training as the classifier of image recognition
Improve the accuracy rate of image recognition.
According to embodiments of the present invention, described that the extensive object image data collection is subjected to image training, obtain depth
Convolutional neural networks AlexNet, further includes:
It is handled using the image concentrated to extensive object image data of mode enhancing technology, obtains that treated schemes
Picture;
Treated that image is trained for described, obtains new depth convolutional neural networks AlexNet.
Specifically, the mode enhancing technology is specially one of rotation, inclination, elastic distortion, cosine function enhancing
Or it is several.It should be understood by those skilled in the art that the present invention not merely limit more than enhancement mode, it is any through the invention
Technical solution carry out mode enhancing method fall in the scope of the present invention.
It should be noted that when obtaining depth convolutional neural networks AlexNet, can also will scheme carrying out image training
As carrying out enhancing processing.Enhancing processing mentioned here, can be modal distortion enhancing.Data enhancement methods are wherein used as, are imitated
It penetrates transformation and elasticity distortion is well-known in character recognition.They are used to generate new samples from original sample and extend instruction
Practice collection.By the way that affine displacement field is applied to mode, it can be generated and such as translate, rotate, scale and inclined simple distortion.Bullet
Property distortion be it is a kind of imitation handwriting style variation image transformation.
Fig. 4 shows the schematic diagram that the present invention carries out image enhancement by cosine mode.
As shown in figure 4, by the present invention in that with the mode conversion method with cosine function.Specifically also apply some increasings
Strong method such as rotates, inclination and elastic distortion.Fig. 4 shows the mode example of cosine function enhancing.The quantity of training image increases
31 times of original 5k training sample are added.By using cosine function, raw mode is left-justify, Right Aligns, top alignment or bottom
Alignment, mode be also center alignment and it is widened.
Fig. 5 shows the schematic diagram that the present invention carries out image enhancement by rotation inclination.
As shown in figure 5, image, is carried out the rotation of certain angle by rotation processing of the present invention by using image;Or
Using the inclined processing of image, purpose is to be transformed to image the image of certain angle is presented with original image.It can also lead to
The mode for crossing elastic distortion carries out image procossing, and original image is transformed to the image of distortion.
By using treated, image is trained, and obtains new depth convolutional neural networks AlexNet.The present invention is logical
Crossing treated, certain distorted image is trained, and can increase the probability of matching image in a network, is reached and is improved image and know
The purpose of other accuracy.
Third aspect present invention provides a kind of computer readable storage medium, includes in the computer readable storage medium
Image processing method program when described image processing method program is executed by processor, realizes a kind of such as above-mentioned image procossing
The step of method.
Image processing method, device and the readable storage medium storing program for executing being related to through the invention, to the minimal characteristic in classifier
Distance is judged, when being greater than preset distance threshold just using classification information as final classification results, can increase figure
As the accuracy of classification.And the convolutional layer in neural network is carried out mentioning for characteristic information by the present invention
It takes, the accuracy of identification can be further increased.The present invention carries out modal distortion increasing also in image training process, for image
By force, the accuracy of image recognition is further increased.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or
It is desirably integrated into another device, or some features can be ignored or not executed.In addition, shown or discussed each composition portion
Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit
Or communication connection, it can be electrical, mechanical or other forms.
Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit
The component shown can be or may not be physical unit;Both it can be located in one place, and may be distributed over multiple network lists
In member;Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In addition, each functional unit in various embodiments of the present invention can be fully integrated in one processing unit, it can also
To be each unit individually as a unit, can also be integrated in one unit with two or more units;It is above-mentioned
Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can store in computer-readable storage medium, which exists
When execution, step including the steps of the foregoing method embodiments is executed;And storage medium above-mentioned includes: movable storage device, read-only deposits
Reservoir (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or
The various media that can store program code such as CD.
If alternatively, the above-mentioned integrated unit of the present invention is realized in the form of software function module and as independent product
When selling or using, it also can store in a computer readable storage medium.Based on this understanding, the present invention is implemented
Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words,
The computer software product is stored in a storage medium, including some instructions are used so that computer equipment (can be with
It is personal computer, server or network equipment etc.) execute all or part of each embodiment the method for the present invention.
And storage medium above-mentioned includes: that movable storage device, ROM, RAM, magnetic or disk etc. are various can store program code
Medium.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of image processing method characterized by comprising
Obtain image information;
According to described image information, feature extraction is carried out using feature extractor, obtains characteristic information;
The characteristic information is sent to classifier;
Classifier is calculated according to the characteristic information, obtains characteristic information and the minimum of hyperplane in corresponding classifier is special
Levy distance;
Judge whether the minimal characteristic distance is greater than preset distance threshold;
If more than then using the classification information of the classifier as final classification information.
2. a kind of image processing method according to claim 1, which is characterized in that the feature extractor is by preparatory
The feature extractor of training setting, specifically:
Obtain extensive object image data collection;
The extensive object image data collection is subjected to image training, obtains depth convolutional neural networks AlexNet;
Obtain the convolutional layer of the depth convolutional neural networks AlexNet;
Using the convolutional layer as feature extractor.
3. a kind of image processing method according to claim 2, it is characterised in that:
The first layer of depth convolutional neural networks AlexNet includes the filter for capturing edge feature, the depth convolution
The middle layer of neural network AlexNet includes multiple convolutional layers and maximum pond layer, the depth convolutional neural networks AlexNet's
The last layer is classification layer, and the weight of each layer of the depth convolutional neural networks AlexNet is determined by training.
4. a kind of image processing method according to claim 2, it is characterised in that:
Feature extractor is used as using the 17th layer of the depth convolutional neural networks AlexNet, it will be in the depth convolution
The 17th layer of obtained characteristic information of neural network AlexNet is sent in classifier.
5. a kind of image processing method according to claim 1, which is characterized in that the classifier is by training in advance
The classifier of setting.
6. a kind of image processing method according to claim 2, which is characterized in that described by the extensive subject image
Data set carries out image training, obtains depth convolutional neural networks AlexNet, further includes:
It is handled using the image concentrated to extensive object image data of mode enhancing technology, the image that obtains that treated;
Treated that image is trained for described, new depth convolutional neural networks AlexNet is obtained, by affiliated new depth
Convolutional neural networks AlexNet convolutional layer is spent as feature extractor.
7. a kind of image processing apparatus, which is characterized in that the device includes: memory, processor, includes figure in the memory
As processing method program, described image processing method program realizes following steps when being executed by the processor:
Obtain image information;
According to described image information, feature extraction is carried out using feature extractor, obtains characteristic information;
The characteristic information is sent to classifier;
Classifier is calculated according to the characteristic information, obtains characteristic information and the minimum of hyperplane in corresponding classifier is special
Levy distance;
Judge whether the minimal characteristic distance is greater than preset distance threshold;
If more than then using the classification information of the classifier as final classification information.
8. a kind of image processing apparatus according to claim 7, which is characterized in that the feature extractor is by preparatory
The feature extractor of training setting, specifically:
Obtain extensive object image data collection;
The extensive object image data collection is subjected to image training, obtains depth convolutional neural networks AlexNet;
Obtain the convolutional layer of the depth convolutional neural networks AlexNet;
Using the convolutional layer as feature extractor.
9. a kind of image processing apparatus according to claim 8, it is characterised in that:
The first layer of depth convolutional neural networks AlexNet includes the filter for capturing edge feature, the depth convolution
The middle layer of neural network AlexNet includes multiple convolutional layers and maximum pond layer, the depth convolutional neural networks AlexNet's
The last layer is classification layer, and the weight of each layer of the depth convolutional neural networks AlexNet is determined by training.
10. a kind of computer readable storage medium, which is characterized in that include image procossing in the computer readable storage medium
Method program when described image processing method program is executed by processor, is realized as described in any one of claims 1 to 6
A kind of the step of image processing method.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910114848.XA CN109993201B (en) | 2019-02-14 | 2019-02-14 | Image processing method, device and readable storage medium |
PCT/CN2019/118277 WO2020164278A1 (en) | 2019-02-14 | 2019-11-14 | Image processing method and device, electronic equipment and readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910114848.XA CN109993201B (en) | 2019-02-14 | 2019-02-14 | Image processing method, device and readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109993201A true CN109993201A (en) | 2019-07-09 |
CN109993201B CN109993201B (en) | 2024-07-16 |
Family
ID=67130148
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910114848.XA Active CN109993201B (en) | 2019-02-14 | 2019-02-14 | Image processing method, device and readable storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109993201B (en) |
WO (1) | WO2020164278A1 (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110570419A (en) * | 2019-09-12 | 2019-12-13 | 杭州依图医疗技术有限公司 | Method and device for acquiring characteristic information and storage medium |
WO2020164278A1 (en) * | 2019-02-14 | 2020-08-20 | 平安科技(深圳)有限公司 | Image processing method and device, electronic equipment and readable storage medium |
CN113239739A (en) * | 2021-04-19 | 2021-08-10 | 深圳市安思疆科技有限公司 | Method and device for identifying wearing article |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112345531B (en) * | 2020-10-19 | 2024-04-09 | 国网安徽省电力有限公司电力科学研究院 | Transformer fault detection method based on bionic robot fish |
CN114639008A (en) * | 2020-12-01 | 2022-06-17 | 中移(成都)信息通信科技有限公司 | Method, device and equipment for extracting remote sensing image building and computer storage medium |
CN112668449A (en) * | 2020-12-24 | 2021-04-16 | 杭州电子科技大学 | Low-risk landform identification method for outdoor autonomous mobile robot |
CN113129279B (en) * | 2021-04-08 | 2024-04-30 | 合肥工业大学 | Composite insulator bird pecking damage risk level assessment method |
CN113642655B (en) * | 2021-08-18 | 2024-02-13 | 杭州电子科技大学 | Small sample image classification method based on support vector machine and convolutional neural network |
CN113901647A (en) * | 2021-09-24 | 2022-01-07 | 成都飞机工业(集团)有限责任公司 | Part process rule compiling method and device, storage medium and electronic equipment |
CN115713763A (en) * | 2022-11-25 | 2023-02-24 | 青海卓旺智慧信息科技有限公司 | Potato image recognition system based on deep learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106372648A (en) * | 2016-10-20 | 2017-02-01 | 中国海洋大学 | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method |
CN106934339A (en) * | 2017-01-19 | 2017-07-07 | 上海博康智能信息技术有限公司 | A kind of target following, the extracting method of tracking target distinguishing feature and device |
CN109033107A (en) * | 2017-06-09 | 2018-12-18 | 腾讯科技(深圳)有限公司 | Image search method and device, computer equipment and storage medium |
CN109145143A (en) * | 2018-08-03 | 2019-01-04 | 厦门大学 | Sequence constraints hash algorithm in image retrieval |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6826300B2 (en) * | 2001-05-31 | 2004-11-30 | George Mason University | Feature based classification |
CN104361096B (en) * | 2014-11-20 | 2016-02-24 | 合肥工业大学 | The image search method of a kind of feature based rich region set |
CN109993201B (en) * | 2019-02-14 | 2024-07-16 | 平安科技(深圳)有限公司 | Image processing method, device and readable storage medium |
-
2019
- 2019-02-14 CN CN201910114848.XA patent/CN109993201B/en active Active
- 2019-11-14 WO PCT/CN2019/118277 patent/WO2020164278A1/en active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106372648A (en) * | 2016-10-20 | 2017-02-01 | 中国海洋大学 | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method |
CN106934339A (en) * | 2017-01-19 | 2017-07-07 | 上海博康智能信息技术有限公司 | A kind of target following, the extracting method of tracking target distinguishing feature and device |
CN109033107A (en) * | 2017-06-09 | 2018-12-18 | 腾讯科技(深圳)有限公司 | Image search method and device, computer equipment and storage medium |
CN109145143A (en) * | 2018-08-03 | 2019-01-04 | 厦门大学 | Sequence constraints hash algorithm in image retrieval |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020164278A1 (en) * | 2019-02-14 | 2020-08-20 | 平安科技(深圳)有限公司 | Image processing method and device, electronic equipment and readable storage medium |
CN110570419A (en) * | 2019-09-12 | 2019-12-13 | 杭州依图医疗技术有限公司 | Method and device for acquiring characteristic information and storage medium |
CN113239739A (en) * | 2021-04-19 | 2021-08-10 | 深圳市安思疆科技有限公司 | Method and device for identifying wearing article |
Also Published As
Publication number | Publication date |
---|---|
WO2020164278A1 (en) | 2020-08-20 |
CN109993201B (en) | 2024-07-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109993201A (en) | A kind of image processing method, device and readable storage medium storing program for executing | |
CN109255322B (en) | A kind of human face in-vivo detection method and device | |
CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
CN103136504B (en) | Face identification method and device | |
US8165410B2 (en) | Bags of visual context-dependent words for generic visual categorization | |
CN109829467A (en) | Image labeling method, electronic device and non-transient computer-readable storage medium | |
CN109934255B (en) | Model fusion method suitable for classification and identification of delivered objects of beverage bottle recycling machine | |
CN105224951B (en) | A kind of vehicle type classification method and sorter | |
Esmaeili et al. | Fast-at: Fast automatic thumbnail generation using deep neural networks | |
CN113076994B (en) | Open-set domain self-adaptive image classification method and system | |
JP6897749B2 (en) | Learning methods, learning systems, and learning programs | |
CN105809205B (en) | A kind of classification method and its system of high spectrum image | |
CN106228166B (en) | The recognition methods of character picture | |
JP2008159056A (en) | Classification through generative model of feature occurring in image | |
CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
CN102147858A (en) | License plate character identification method | |
CN110222582B (en) | Image processing method and camera | |
CN105095884A (en) | Pedestrian recognition system and pedestrian recognition processing method based on random forest support vector machine | |
CN107368827A (en) | Character identifying method and device, user equipment, server | |
CN116612307A (en) | Solanaceae disease grade identification method based on transfer learning | |
KR101093107B1 (en) | Image information classification method and apparatus | |
CN104966109A (en) | Medical laboratory report image classification method and apparatus | |
CN109741351A (en) | A kind of classification responsive type edge detection method based on deep learning | |
CN105069133B (en) | A kind of digital picture sorting technique based on Unlabeled data | |
CN117058534A (en) | Small sample remote sensing image target detection method based on meta-knowledge adaptive migration network |
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 |