CN112613333A - Method for calculating difference between network output image and label - Google Patents
Method for calculating difference between network output image and label Download PDFInfo
- Publication number
- CN112613333A CN112613333A CN201911377864.4A CN201911377864A CN112613333A CN 112613333 A CN112613333 A CN 112613333A CN 201911377864 A CN201911377864 A CN 201911377864A CN 112613333 A CN112613333 A CN 112613333A
- Authority
- CN
- China
- Prior art keywords
- image
- reference images
- comparison result
- processed
- generated
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of image processing, in particular to a method for calculating the difference between a network output image and a label, aiming at solving the problem that creating a large amount of label data in the prior art is time-consuming and labor-consuming. A plurality of first reference images are obtained, wherein each of the plurality of first reference images comprises object image data corresponding to the target. And generating a target image according to the template label image and the plurality of first reference images, wherein the target image comprises a generated object, the outline of the generated object is generated according to the template label image, and the color or the material of the generated object is generated according to the plurality of first reference images so as to automatically obtain the plausible image with the same distribution as the template label image.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a method for calculating the difference between a network output image and a label.
Background
The remote sensing image is one of important data of spatial information, and is widely applied to the fields of geological and flood disaster monitoring, agricultural and forest resource investigation, land utilization and urban planning and military. With the development of the space science and the earth observation technology in China, the data of the remote sensing image data has an exponential growth trend every year, and the effective management of the mass remote sensing image data becomes increasingly important.
The remote sensing image labeling is one of important contents for analyzing and understanding the remote sensing image, and is realized by extracting bottom visual features of the remote sensing image and learning the relation between the bottom visual features and high-level semantics through some machine learning models, so that some semantic labels are automatically labeled to the remote sensing image. The automatic marking of the remote sensing images is an understanding of the semantics of the remote sensing images and is an important technical basis for category cataloging and searching of mass remote sensing images.
The automatic labeling work of the remote sensing image can be regarded as the generalized automatic classification work of the remote sensing image, namely before the automatic labeling work of the remote sensing image, the category labels (namely text labels) corresponding to the remote sensing image to be labeled need to be determined, and then different remote sensing images are correspondingly linked with different category labels. The labeling work of the traditional image mainly comprises 3 types of methods: object ontology based methods, machine learning based methods and correlation feedback based methods. The traditional image labeling work is mainly to analyze and understand the visual content of the image through the low-level visual features of the image.
However, with the rapid development of machine learning, it is laborious and time-consuming for researchers to create a large amount of tag data. Therefore, how to solve the above problems is very important nowadays.
Disclosure of Invention
Therefore, the technical problem to be solved by the present invention is to overcome the defect that creating a large amount of label data in the prior art is time-consuming and labor-consuming, thereby providing a method for calculating the gap between the network output image and the label.
The technical purpose of the invention is realized by the following technical scheme:
a method for calculating the difference between a network output image and a label comprises the following steps:
obtaining a template label image, wherein the template label image comprises a label corresponding to an object;
obtaining a plurality of first reference images, wherein each of the plurality of first reference images comprises object image data corresponding to the target;
generating a target image according to the template label image and the plurality of first reference images, wherein the target image comprises a generated object, a contour of the generated object is generated according to the template label image, and a color or a material of the generated object is generated according to the plurality of first reference images;
the target image is generated by generating a confrontation network model, and a training data of the generated confrontation network model comprises the template label image and the plurality of first reference images.
Optionally, the method further comprises:
an image processing model is trained via the target image and the template label image, wherein the trained image processing model is used for processing an input image without labels to generate a label image associated with the input image.
Optionally, the method further comprises:
obtaining a background and an object of the input image through the image processing model;
the tag image is generated according to the background and the object, wherein the tag image comprises a first tag associated with the object and a second tag associated with the background.
Optionally, before the operation of generating the target image, the image processing method further comprises:
training an image generation engine, wherein the image generation engine is used for generating the target image.
Optionally, the operation of training the image generation engine comprises:
generating a processed image according to the template label image and the plurality of first reference images;
comparing the processed image with the plurality of first reference images;
in response to whether a comparison result is higher than a threshold value, the image generation engine is updated or the training of the image generation engine is suspended.
Optionally, the method further comprises:
in response to the comparison result being higher than the threshold value, updating the processed image according to the comparison result, and comparing the processed image with the plurality of first reference images until the comparison result is lower than the threshold value;
and terminating training the image generation engine in response to the comparison result being less than the threshold value.
Optionally, the operation of comparing the processed image with the plurality of first reference images comprises:
comparing a color, a texture or a content object shape of the processed image and the plurality of first reference images.
Optionally, the operation of training the image generation engine comprises:
generating a processed image according to the template label image and the plurality of first reference images;
generating a generation background and a generation object based on the processed image;
forming a processing foreground image according to the generated object;
obtaining a plurality of second reference images, wherein each of the plurality of second reference images comprises first object image data corresponding to a color of the target and first background image data having a single color;
comparing the processed foreground image with the plurality of second reference images to obtain a first comparison result;
updating the processed image according to whether the first comparison result is higher than a critical value.
Optionally, the operation of training the image generation engine further comprises:
forming a processed background image according to the generated background;
obtaining a plurality of third reference images, wherein each of the plurality of third reference images comprises second object image data corresponding to the target and having a single color and second background image data having a color;
comparing the processed background image with the plurality of third reference images to obtain a second comparison result;
updating the processed image according to whether the second comparison result is higher than the threshold value.
Optionally, the operation of training the image generation engine further comprises:
comparing the processed image with the plurality of first reference images to obtain a third comparison result;
in response to the third comparison result being higher than the threshold value, updating the processed image according to the third comparison result;
and stopping training the image generation engine according to the condition that the first comparison result, the second comparison result and the third comparison result are all lower than the critical value.
According to the technical scheme of the invention, by using the method, a large number of pixel level images with labels can be automatically generated, so that high accuracy can be obtained when the operation of dividing the object or dividing the object from the image is executed.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
A method for calculating the difference between a network output image and a label comprises the following steps:
obtaining a template label image, wherein the template label image comprises a label corresponding to an object;
obtaining a plurality of first reference images, wherein each of the plurality of first reference images comprises object image data corresponding to the target;
generating a target image according to the template label image and the plurality of first reference images, wherein the target image comprises a generated object, a contour of the generated object is generated according to the template label image, and a color or a material of the generated object is generated according to the plurality of first reference images;
the target image is generated by generating a confrontation network model, and a training data of the generated confrontation network model comprises the template label image and the plurality of first reference images.
In some embodiments, the method further comprises:
an image processing model is trained via the target image and the template label image, wherein the trained image processing model is used for processing an input image without labels to generate a label image associated with the input image.
In some embodiments, the method further comprises:
obtaining a background and an object of the input image through the image processing model;
the tag image is generated according to the background and the object, wherein the tag image comprises a first tag associated with the object and a second tag associated with the background.
Wherein the image processing method further comprises, before the operation of generating the target image:
training an image generation engine, wherein the image generation engine is used for generating the target image.
Wherein training the image generation engine comprises:
generating a processed image according to the template label image and the plurality of first reference images;
comparing the processed image with the plurality of first reference images;
in response to whether a comparison result is higher than a threshold value, the image generation engine is updated or the training of the image generation engine is suspended.
In some embodiments, the method further comprises:
in response to the comparison result being higher than the threshold value, updating the processed image according to the comparison result, and comparing the processed image with the plurality of first reference images until the comparison result is lower than the threshold value;
and terminating training the image generation engine in response to the comparison result being less than the threshold value.
Wherein comparing the processed image to the plurality of first reference images comprises:
comparing a color, a texture or a content object shape of the processed image and the plurality of first reference images.
In some embodiments, the training the image generation engine includes:
generating a processed image according to the template label image and the plurality of first reference images;
generating a generation background and a generation object based on the processed image;
forming a processing foreground image according to the generated object;
obtaining a plurality of second reference images, wherein each of the plurality of second reference images comprises first object image data corresponding to a color of the target and first background image data having a single color;
comparing the processed foreground image with the plurality of second reference images to obtain a first comparison result;
updating the processed image according to whether the first comparison result is higher than a critical value.
Wherein training the image generation engine further comprises:
forming a processed background image according to the generated background;
obtaining a plurality of third reference images, wherein each of the plurality of third reference images comprises second object image data corresponding to the target and having a single color and second background image data having a color;
comparing the processed background image with the plurality of third reference images to obtain a second comparison result;
updating the processed image according to whether the second comparison result is higher than the threshold value.
Wherein training the image generation engine further comprises:
comparing the processed image with the plurality of first reference images to obtain a third comparison result;
in response to the third comparison result being higher than the threshold value, updating the processed image according to the third comparison result;
and stopping training the image generation engine according to the condition that the first comparison result, the second comparison result and the third comparison result are all lower than the critical value.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications therefrom are within the scope of the invention.
Claims (10)
1. A method for calculating the difference between a network output image and a label is characterized by comprising the following steps:
obtaining a template label image, wherein the template label image comprises a label corresponding to an object;
obtaining a plurality of first reference images, wherein each of the plurality of first reference images comprises object image data corresponding to the target;
generating a target image according to the template label image and the plurality of first reference images, wherein the target image comprises a generated object, a contour of the generated object is generated according to the template label image, and a color or a material of the generated object is generated according to the plurality of first reference images;
the target image is generated by generating a confrontation network model, and a training data of the generated confrontation network model comprises the template label image and the plurality of first reference images.
2. The method of claim 1, further comprising:
an image processing model is trained via the target image and the template label image, wherein the trained image processing model is used for processing an input image without labels to generate a label image associated with the input image.
3. The method of claim 1, further comprising:
obtaining a background and an object of the input image through the image processing model;
the tag image is generated according to the background and the object, wherein the tag image comprises a first tag associated with the object and a second tag associated with the background.
4. The method of claim 1, wherein the image processing method further comprises, before the operation of generating the target image:
training an image generation engine, wherein the image generation engine is used for generating the target image.
5. The method of claim 4, wherein the operation of training the image generation engine comprises:
generating a processed image according to the template label image and the plurality of first reference images;
comparing the processed image with the plurality of first reference images;
in response to whether a comparison result is higher than a threshold value, the image generation engine is updated or the training of the image generation engine is suspended.
6. The method of claim 5, further comprising:
in response to the comparison result being higher than the threshold value, updating the processed image according to the comparison result, and comparing the processed image with the plurality of first reference images until the comparison result is lower than the threshold value;
and terminating training the image generation engine in response to the comparison result being less than the threshold value.
7. The method of claim 5, wherein comparing the processed image with the plurality of first reference images comprises:
comparing a color, a texture or a content object shape of the processed image and the plurality of first reference images.
8. The method of claim 4, wherein the operation of training the image generation engine comprises:
generating a processed image according to the template label image and the plurality of first reference images;
generating a generation background and a generation object based on the processed image;
forming a processing foreground image according to the generated object;
obtaining a plurality of second reference images, wherein each of the plurality of second reference images comprises first object image data corresponding to a color of the target and first background image data having a single color;
comparing the processed foreground image with the plurality of second reference images to obtain a first comparison result;
updating the processed image according to whether the first comparison result is higher than a critical value.
9. The method of claim 8, wherein the operation of training the image generation engine further comprises:
forming a processed background image according to the generated background;
obtaining a plurality of third reference images, wherein each of the plurality of third reference images comprises second object image data corresponding to the target and having a single color and second background image data having a color;
comparing the processed background image with the plurality of third reference images to obtain a second comparison result;
updating the processed image according to whether the second comparison result is higher than the threshold value.
10. The method of claim 9, wherein the operation of training the image generation engine further comprises:
comparing the processed image with the plurality of first reference images to obtain a third comparison result;
in response to the third comparison result being higher than the threshold value, updating the processed image according to the third comparison result;
and stopping training the image generation engine according to the condition that the first comparison result, the second comparison result and the third comparison result are all lower than the critical value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911377864.4A CN112613333A (en) | 2019-12-27 | 2019-12-27 | Method for calculating difference between network output image and label |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911377864.4A CN112613333A (en) | 2019-12-27 | 2019-12-27 | Method for calculating difference between network output image and label |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112613333A true CN112613333A (en) | 2021-04-06 |
Family
ID=75225672
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911377864.4A Withdrawn CN112613333A (en) | 2019-12-27 | 2019-12-27 | Method for calculating difference between network output image and label |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112613333A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805169A (en) * | 2017-05-04 | 2018-11-13 | 宏达国际电子股份有限公司 | Image treatment method, non-transient computer readable media and image processing system |
| CN110347857A (en) * | 2019-06-06 | 2019-10-18 | 武汉理工大学 | The semanteme marking method of remote sensing image based on intensified learning |
-
2019
- 2019-12-27 CN CN201911377864.4A patent/CN112613333A/en not_active Withdrawn
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805169A (en) * | 2017-05-04 | 2018-11-13 | 宏达国际电子股份有限公司 | Image treatment method, non-transient computer readable media and image processing system |
| CN110347857A (en) * | 2019-06-06 | 2019-10-18 | 武汉理工大学 | The semanteme marking method of remote sensing image based on intensified learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109741332B (en) | Man-machine cooperative image segmentation and annotation method | |
| US10719301B1 (en) | Development environment for machine learning media models | |
| US11537506B1 (en) | System for visually diagnosing machine learning models | |
| CN112241452B (en) | Model training method and device, electronic equipment and storage medium | |
| CN110399800B (en) | License plate detection method and system based on deep learning VGG16 framework and storage medium | |
| CN105678309B (en) | A kind of image multi-tag dimensioning algorithm based on more example packet feature learnings | |
| CN108229578B (en) | Image data target identification method based on three layers of data, information and knowledge map framework | |
| CN108241867B (en) | Classification method and device | |
| CN105005760A (en) | Pedestrian re-identification method based on finite mixture model | |
| CN115048316B (en) | Semi-supervised software code defect detection method and device | |
| CN110912917A (en) | Malicious URL detection method and system | |
| CN113128565B (en) | Automatic image annotation system and device oriented to agnostic pre-training annotation data | |
| CN114491071A (en) | Food safety knowledge graph construction method and system based on cross-media data | |
| CN109857892A (en) | Semi-supervised cross-module state Hash search method based on category transmitting | |
| CN117611988A (en) | Automatic identification and monitoring method and system for newly-increased farmland management and protection attribute | |
| CN112613333A (en) | Method for calculating difference between network output image and label | |
| US20230260262A1 (en) | Automated annotation of visual data through computer vision template matching | |
| Yiwen et al. | Prediction and counting of field wheat based on lc-dcvgg | |
| CN115169578A (en) | AI model production method and system based on meta-space data markers | |
| CN113255722A (en) | Image annotation method and device | |
| CN113920450A (en) | Method and device for identifying insulator RTV coating based on intrinsic image decomposition | |
| Ivanovska et al. | Tomatodiff: On–plant tomato segmentation with denoising diffusion models | |
| JP2017201483A (en) | Table motif extraction device, classifier learning device, table type classifier, method, and program | |
| Xu et al. | Generating visual and semantic explanations with multi-task network | |
| CN112199287A (en) | Cross-project software defect prediction method based on enhanced hybrid expert model |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210406 |
|
| WW01 | Invention patent application withdrawn after publication |