CN109658349A - A kind of image enchancing method and its application for supervised learning application - Google Patents
A kind of image enchancing method and its application for supervised learning application Download PDFInfo
- Publication number
- CN109658349A CN109658349A CN201811367395.3A CN201811367395A CN109658349A CN 109658349 A CN109658349 A CN 109658349A CN 201811367395 A CN201811367395 A CN 201811367395A CN 109658349 A CN109658349 A CN 109658349A
- Authority
- CN
- China
- Prior art keywords
- data
- image
- supervised learning
- enhancing
- learning application
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 230000002708 enhancing effect Effects 0.000 claims abstract description 56
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000012216 screening Methods 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims description 16
- 238000012549 training Methods 0.000 abstract description 17
- 230000001537 neural effect Effects 0.000 abstract 1
- 238000005516 engineering process Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 241001062009 Indigofera Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of image enchancing method for supervised learning application and its applications, 1) Enhancement Method is the following steps are included: obtain original data set to be reinforced；2) according to supervised learning application task, setting data enhances parameter；3) respectively to the image data and label data progress data enhancing processing in original data set；4) enhanced image and markup information are screened；5) by after screening image and markup information be stored into original data set.Compared with prior art, the present invention has many advantages, such as effectively obtain the data for meeting supervised learning data set data distribution, improves neural metwork training precision.
Description
Technical field
The present invention relates to depth learning technology fields, more particularly, to a kind of image enhancement side for supervised learning application
Method and its application.
Background technique
With the extensive use of multimedia technology and computer network, on electronic equipment and there is great amount of images number on network
According to.In order to effectively manage or search for these image files, for user provide preferably experience service, to these pictures into
It is more and more important that row identification and segmentation etc. obtain pictorial information in turn.
Supervised learning is machine learning method and important one of the learning method of deep learning algorithm field.Supervised learning is
By existing training sample (i.e. given data and its corresponding output) Lai Xunlian, so that an optimal models are obtained, then
Using this model by all new data samples be mapped as accordingly export as a result, to output result carry out simply judge from
And realize the purpose of classification, then this optimal models is also just provided with the ability classified to unknown data.Supervised learning
As long as input sample collection in, machine can therefrom deduce out the possible outcome for the variable that sets objectives.By being carried out to training set
Supervised learning, and test set is predicted, to achieve the purpose that prediction.Supervised learning in the training process, in order to allow mind
Feature through the more preferable learning training data of network needs to input a large amount of training data with lift scheme effect.This there is
Following two problems: first, if it is to carry out the study for having supervision to neural network model, then being inputted by training data
Before neural network, it is necessary to manually mark corresponding label to each training data；Once the amount of training data is larger, need
The cost of labor to be paid also will be bigger；Second, often there is training data deficiency in reality.
To solve the above-mentioned problems, at present usually using the method to training data progress data enhancing to realize to training
The expansion of data volume.But in current data enhancement methods, there are problems that following two, first, being typically all for picture
Data enhancing is carried out, not doing to the label of training data mark enhances same transformation with data；Second, in use to training
Data execute when being formed by enhancing data training neural network model after data enhancement operations, the often essence of neural network model
It spends lower.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide one kind to be directed to supervised learning
The image enchancing method of application and its application, can be when carrying out data enhancement operations to supervised learning data set, more preferably
Ground generates some data for meeting supervised learning data set data distribution.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of image enchancing method for supervised learning application, method includes the following steps:
1) original data set to be reinforced is obtained；
2) according to supervised learning application task, setting data enhances parameter；
3) respectively to the image data and label data progress data enhancing processing in original data set；
4) enhanced image and markup information are screened；
5) by after screening image and markup information be stored into original data set.
Further, in the step 2), data enhancing parameter includes data enhancing type, each data enhancing applied probability
Enhance threshold value with data.
Further, the data enhancing type includes the combination of a kind of data enhancing or the enhancing of a variety of data.
Further, every kind of data are enhanced, the data enhancing threshold value is random within the set range.
Further, in the step 3), the data enhancing processing to image data includes pixel color transformation and space
Geometric transformation；
Data enhancing processing to label data only includes that space geometry converts.
Further, in the step 4), increased according to pixel number of the enhanced label data in figure and label data
The ratio between pixel number after strong and the pixel number before enhancing are screened.
Further, it in the step 4), screens enhanced image and markup information includes:
A the label data not being inconsistent with former data distribution) is deleted；
B the image of no corresponding label data) is deleted.
The present invention also provides a kind of supervised learning methods based on image enchancing method as mentioned.
Compared with prior art, the present invention have with following the utility model has the advantages that
(1) labeled data and picture of data set are done data enhancing simultaneously by the present invention, effectively improve supervised learning application
When available training data.
(2) present invention sets the relevant parameter of data enhancing according to supervised learning application task, is more in line with reality,
It is more convenient to use, it effectively improves using precision.
(3) present invention enhances every kind of data, the parameter values of enhancing be it is random in the threshold value of setting, with control
The enhanced data space of data is in controllable range.
(4) present invention screens enhanced data, deletes the data enhancing that may destroy data space as a result, having
Effect improves the training precision of neural network model.
Detailed description of the invention
Fig. 1 is flow diagram of the invention.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to
Following embodiments.
As shown in Figure 1, the present invention provides a kind of image enchancing method for supervised learning application, this method includes following
Step:
1) original data set to be reinforced is obtained.Original data set includes picture and its markup information for corresponding to task.
2) according to supervised learning application task, setting data enhances parameter, including data enhancing type, the enhancing of each data are answered
Enhance threshold value with probability and data.
It for different supervised learning tasks, should be also different using the mode that data enhance, such as be directed to the knowledge of face
Other task, data enhancement method is overturning, brightness, contrast are best；For the Building recognition taken photo by plane, data enhancement method is
Rotation, changes having a size of best interception.It is for the requirement of data space of pending data enhancing in different supervised learning tasks
It is different, in order to meet this requirement, following three different strategies have been done for the mode of determining data enhancing to meet
For the being consistent property of data space of data enhancing front and back: selecting a kind of enhancing in the enhancing of a variety of data first, can apply
Or successively a variety of data of application enhance；Second, can choose the applied probability of each data enhancing；Third, for every kind of number
According to enhancing, threshold value be also it is random in the range of setting, to control the enhanced data space of data in controllable range.
3) respectively to the image data and label data progress data enhancing processing in original data set.
For image enhancing there are many kinds of mode, be broadly divided into pixel color transformation classes and space geometry transformation classes.
For each different supervised learning task, different image enhancement modes selects all different.
For label data, pixel color variation is without transformation, but space geometry transformation needs to do phase according to different task
It should convert.Such as dividing task, label data is a target one figure, in space geometry transformation, just by number of tags
It is converted simultaneously according to as image with image；For identification mission, label data is one tandem List (target of a target
Coordinate information), space geometry transformation when, label data information needs to do individual transformation.
4) enhanced image and markup information are screened.
According to the needs of supervised learning task, increased according to pixel number of the enhanced label data in figure and label data
The ratio between pixel number after strong and the pixel number before enhancing are screened, to delete some possible data enhancings for destroying data space
Mode.Screening includes: A) delete the label data not being inconsistent with former data distribution；B the image of no corresponding label data) is deleted.
5) by after screening image and markup information be stored into original data set.
The present invention can realize more effective supervised learning method based on above-mentioned image enchancing method.
The present embodiment is to identify the character on automotive license plate (text, letter and number) for supervised learning task, in utilization
The realization process for stating method progress image enhancement is as follows.
Step 1: image and markup information to be reinforced are obtained
The image of the present embodiment is to take pictures to obtain in various scenes from all angles to different automotive license plates, is then used
Read in image data in the library OPENCV.Markup information be using annotation tool according on each figure by the character one by one of license plate
What (text, letter and number) mark obtained, image data and markup information are then stored into number using dictionary Container Type
According to.
Step 2: data enhancing is carried out to image to be reinforced and markup information
A) mode of selection data enhancing
For Recognition of License Plate Characters task, can successively it enhance using a variety of data, the applied probability of each data enhancing
It is 50%.In society, the picture scene for shooting license plate is clapped on road, so the influence in addition to illumination can ratio
Other than larger, other transformation will not be especially big.So, in addition to the data enhancing of brightness, the threshold value of other data enhancings is set in
It is random in one smaller range.
B) data enhancing is carried out to image data
For Recognition of License Plate Characters task, since the color of license plate is basically unchanged, in addition to the influence of illumination, license plate indigo plant bottom is white
Word is constant, so in pixel color transformation classes, the method for these types of data enhancing is proper: random brightness and random
Contrast；For alphanumeric identification, I and 1,6 and 9 is easy to obscure, then rotate, overturn it is just very unsuitable, so in sky
Between in transformation classes, the method for these types of data enhancing is proper: random scale and random interception.The parameter of image data enhancing
As shown in table 1.
Table 1
C) data enhancing is carried out to label data
For Recognition of License Plate Characters task, label data is tandem List of character (text, number, letter)
(coordinate information of target), in pixel color transformation, label data does not need to convert, but converts in space geometry, array
Information needs to do individual transformation.
Step 3: screening enhancing image and markup information
The character of enhanced license plate in image border, may be set in figure due to intercepting for the character recognition of license plate
The ratio between pixel number before pixel number and enhancing is necessarily equal to 1, that is, must character must all be shown in figure, Cai Nengna
Enter in data set.If there is no the character on any license plate in image, this image is just deleted.
Step 4: preservation has enhanced image and markup information
The image enhanced and markup information are saved into original data set.
The preferred embodiment of the present invention has been described in detail above.It should be appreciated that those skilled in the art without
It needs creative work according to the present invention can conceive and makes many modifications and variations.Therefore, all technologies in the art
Personnel are available by logical analysis, reasoning, or a limited experiment on the basis of existing technology under this invention's idea
Technical solution, all should be within the scope of protection determined by the claims.
Claims (8)
1. a kind of image enchancing method for supervised learning application, which is characterized in that method includes the following steps:
1) original data set to be reinforced is obtained；
2) according to supervised learning application task, setting data enhances parameter；
3) respectively to the image data and label data progress data enhancing processing in original data set；
4) enhanced image and markup information are screened；
5) by after screening image and markup information be stored into original data set.
2. the image enchancing method according to claim 1 for supervised learning application, which is characterized in that the step 2)
In, data enhancing parameter includes data enhancing type, each data enhancing applied probability and data enhancing threshold value.
3. the image enchancing method according to claim 2 for supervised learning application, which is characterized in that the data increase
Strong type includes the combination of a kind of data enhancing or the enhancing of a variety of data.
4. the image enchancing method according to claim 2 for supervised learning application, which is characterized in that every kind of data
Enhancing, the data enhancing threshold value are random within the set range.
5. the image enchancing method according to claim 1 for supervised learning application, which is characterized in that the step 3)
In, the data enhancing processing to image data includes that pixel color transformation and space geometry convert；
Data enhancing processing to label data only includes that space geometry converts.
6. the image enchancing method according to claim 1 for supervised learning application, which is characterized in that the step 4)
In, according to pixel number of the enhanced label data in figure and the pixel number before the enhanced pixel number of label data and enhancing
The ratio between screened.
7. the image enchancing method according to claim 6 for supervised learning application, which is characterized in that the step 4)
In, it screens enhanced image and markup information includes:
A the label data not being inconsistent with former data distribution) is deleted；
B the image of no corresponding label data) is deleted.
8. a kind of supervised learning method based on the image enchancing method as described in claim 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811367395.3A CN109658349A (en) | 2018-11-16 | 2018-11-16 | A kind of image enchancing method and its application for supervised learning application |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811367395.3A CN109658349A (en) | 2018-11-16 | 2018-11-16 | A kind of image enchancing method and its application for supervised learning application |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109658349A true CN109658349A (en) | 2019-04-19 |
Family
ID=66110976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811367395.3A Pending CN109658349A (en) | 2018-11-16 | 2018-11-16 | A kind of image enchancing method and its application for supervised learning application |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109658349A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598799A (en) * | 2020-04-30 | 2020-08-28 | 中国科学院深圳先进技术研究院 | Image toning enhancement method and image toning enhancement neural network training method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330453A (en) * | 2017-06-19 | 2017-11-07 | 中国传媒大学 | The Pornographic image recognizing method of key position detection is recognized and merged based on substep |
| EP3246875A2 (en) * | 2016-05-18 | 2017-11-22 | Siemens Healthcare GmbH | Method and system for image registration using an intelligent artificial agent |
| CN107958271A (en) * | 2017-12-06 | 2018-04-24 | 电子科技大学 | The cutaneous lesions deep learning identifying system of Analysis On Multi-scale Features based on expansion convolution |
| CN108230233A (en) * | 2017-05-16 | 2018-06-29 | 北京市商汤科技开发有限公司 | Data enhancing, treating method and apparatus, electronic equipment and computer storage media |
| CN108446617A (en) * | 2018-03-09 | 2018-08-24 | 华南理工大学 | The human face quick detection method of anti-side face interference |
-
2018
- 2018-11-16 CN CN201811367395.3A patent/CN109658349A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3246875A2 (en) * | 2016-05-18 | 2017-11-22 | Siemens Healthcare GmbH | Method and system for image registration using an intelligent artificial agent |
| CN108230233A (en) * | 2017-05-16 | 2018-06-29 | 北京市商汤科技开发有限公司 | Data enhancing, treating method and apparatus, electronic equipment and computer storage media |
| CN107330453A (en) * | 2017-06-19 | 2017-11-07 | 中国传媒大学 | The Pornographic image recognizing method of key position detection is recognized and merged based on substep |
| CN107958271A (en) * | 2017-12-06 | 2018-04-24 | 电子科技大学 | The cutaneous lesions deep learning identifying system of Analysis On Multi-scale Features based on expansion convolution |
| CN108446617A (en) * | 2018-03-09 | 2018-08-24 | 华南理工大学 | The human face quick detection method of anti-side face interference |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598799A (en) * | 2020-04-30 | 2020-08-28 | 中国科学院深圳先进技术研究院 | Image toning enhancement method and image toning enhancement neural network training method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | DMA-Net: DeepLab with multi-scale attention for pavement crack segmentation | |
| Ren et al. | gSLICr: SLIC superpixels at over 250Hz | |
| Thilagamani et al. | Gaussian and gabor filter approach for object segmentation | |
| Pepik et al. | What is holding back convnets for detection? | |
| US20210224969A1 (en) | Detection and identification of objects in images | |
| CN108230233A (en) | Data enhancing, treating method and apparatus, electronic equipment and computer storage media | |
| CN113822951B (en) | Image processing method, device, electronic equipment and storage medium | |
| US10810433B2 (en) | Method and system for tracking objects | |
| Wang et al. | Semantic segmentation of remote sensing ship image via a convolutional neural networks model | |
| Xiang et al. | Lightweight fully convolutional network for license plate detection | |
| Lu et al. | License plate detection and recognition using hierarchical feature layers from CNN | |
| CN110991439A (en) | Method for extracting handwritten characters based on pixel-level multi-feature joint classification | |
| CN110956167A (en) | Classification discrimination and strengthened separation method based on positioning characters | |
| Mukherjee et al. | Object detection under challenging lighting conditions using high dynamic range imagery | |
| Rio-Alvarez et al. | Effects of Challenging Weather and Illumination on Learning‐Based License Plate Detection in Noncontrolled Environments | |
| Jin et al. | Vehicle license plate recognition for fog‐haze environments | |
| CN115082551A (en) | Multi-target detection method based on unmanned aerial vehicle aerial video | |
| Manasa Devi et al. | Detection and tracking of text from video using MSER and SIFT | |
| CN109658349A (en) | A kind of image enchancing method and its application for supervised learning application | |
| Grard et al. | Object segmentation in depth maps with one user click and a synthetically trained fully convolutional network | |
| Rafi et al. | Salient object detection employing regional principal color and texture cues | |
| Hezil et al. | Vehicle license plate detection using morphological operations and deep learning | |
| Kumdakcı et al. | Generative data augmentation for vehicle detection in aerial images | |
| Jeong et al. | Homogeneity patch search method for voting-based efficient vehicle color classification using front-of-vehicle image | |
| Masita et al. | Refining the efficiency of R-CNN in Pedestrian Detection |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190419 |
|
| RJ01 | Rejection of invention patent application after publication |