CN-121982299-A - Image labeling method, device, computer equipment and storage medium

CN121982299ACN 121982299 ACN121982299 ACN 121982299ACN-121982299-A

Abstract

The invention discloses an image labeling method, an image labeling device, computer equipment and a storage medium. The method comprises the steps of carrying out mask processing on an image to be marked, determining a plurality of segmentation masks corresponding to the image to be marked and segmentation scores corresponding to each segmentation mask, carrying out significance prior analysis on the plurality of segmentation masks corresponding to the image to be marked, determining target prior scores corresponding to the plurality of segmentation masks, determining comprehensive significance scores corresponding to the plurality of segmentation masks based on the segmentation scores corresponding to the plurality of segmentation masks and the target prior scores, screening the plurality of segmentation masks based on the comprehensive significance scores corresponding to the plurality of segmentation masks, determining a plurality of target masks corresponding to the image to be marked, and marking the image to be marked based on the plurality of target masks corresponding to the image to be marked. The method can achieve the purpose of high-quality and reliable labeling of the image to be labeled, does not need manual participation, and can achieve the purpose of high-efficiency labeling.

Inventors

LIU WENLONG
WEI XINMING
YU XIAOTIAN
LI AIJUN

Assignees

深圳云天励飞技术股份有限公司

Dates

Publication Date: 20260505
Application Date: 20251219

Claims (10)

1. An image labeling method, comprising: Carrying out mask processing on an image to be marked, and determining a plurality of segmentation masks corresponding to the image to be marked and segmentation scores corresponding to each segmentation mask; performing significance prior analysis on a plurality of segmentation masks corresponding to the image to be marked, and determining target prior scores corresponding to the plurality of segmentation masks; Determining a comprehensive significance score corresponding to the plurality of segmentation masks based on the segmentation scores corresponding to the plurality of segmentation masks and the target prior score; Screening a plurality of segmentation masks based on comprehensive significance scores corresponding to the segmentation masks, and determining a plurality of target masks corresponding to the image to be marked; and labeling the image to be labeled based on a plurality of target masks corresponding to the image to be labeled.
2. The image labeling method according to claim 1, wherein the masking the image to be labeled, determining a plurality of segmentation masks corresponding to the image to be labeled and a segmentation score corresponding to each segmentation mask, comprises: performing prompt point analysis on the image to be marked to determine a prompt point set corresponding to the image to be marked, wherein the prompt point set comprises a plurality of target prompt points; And carrying out segmentation mask processing on the image to be marked based on the prompt point set corresponding to the image to be marked, and determining the segmentation mask and the segmentation score corresponding to each target prompt point.
3. The method for labeling an image according to claim 2, wherein the performing a prompt point analysis on the image to be labeled to determine a prompt point set corresponding to the image to be labeled includes: generating grids based on the image to be annotated, and determining a plurality of uniform grid points; extracting a saliency priori region of the image to be marked, and determining a saliency priori graph, and a foreground region and a background region corresponding to the saliency priori graph; Performing cluster analysis on a plurality of uniform grid points of the foreground region to determine foreground prompt points corresponding to the foreground region; Performing cluster analysis on a plurality of uniform grid points of the background area to determine background prompting points corresponding to the foreground area; And determining a prompt point set corresponding to the image to be marked based on the uniform grid points, the foreground prompt points and the background prompt points.
4. The method according to claim 3, wherein the performing a saliency prior analysis on the plurality of segmentation masks corresponding to the image to be annotated to determine the target prior scores corresponding to the plurality of segmentation masks includes: Filtering a plurality of segmentation masks corresponding to the image to be annotated, and determining a plurality of effective segmentation masks corresponding to the image to be annotated; And performing significance prior analysis on the multiple effective segmentation masks corresponding to the image to be marked, and determining target prior scores corresponding to the multiple effective segmentation masks.
5. The method according to claim 4, wherein the filtering the plurality of segmentation masks corresponding to the image to be annotated to determine the plurality of effective segmentation masks corresponding to the image to be annotated includes: performing overlapping degree analysis on the segmentation masks corresponding to any two foreground prompt points, determining a first merging overlapping degree, merging the segmentation masks corresponding to the two foreground prompt points with the first merging overlapping degree larger than a first preset overlapping degree threshold value, and forming a foreground merging mask; Performing overlapping degree analysis on the segmentation masks corresponding to any two background prompting points, determining second merging overlapping degree, merging the segmentation masks corresponding to the two background prompting points with the second merging overlapping degree larger than a second preset overlapping degree threshold value, and forming a background merging mask; Performing overlapping degree analysis on the segmentation masks corresponding to the uniform grid points and the foreground merging masks to determine foreground overlapping degree, and performing overlapping degree analysis on the segmentation masks corresponding to the uniform grid points and the background merging masks to determine background overlapping degree; Determining a segmentation mask corresponding to uniform grid points meeting preset filtering conditions as a grid point mask, wherein the preset filtering conditions comprise that the foreground overlapping degree is larger than a third preset overlapping degree threshold value and the background overlapping degree is smaller than the third preset overlapping degree threshold value; And determining the foreground merging mask and the grid point mask as effective segmentation masks corresponding to the images to be annotated.
6. The image labeling method of claim 1, wherein the target prior score is determined based on a center prior score, a boundary connected prior score, a color contrast prior score, a size prior score, and a saliency prior graph goodness-of-fit score; The center prior score is determined based on a distance between a centroid of the segmentation mask and an image center of the image to be annotated; the boundary communication priori score is determined based on the contact proportion between the segmentation mask and the image boundary of the image to be marked; The color contrast prior score is determined based on the contrast of pixels corresponding to the segmentation mask and pixels corresponding to a background area around the segmentation mask in a color space; the size prior score is determined based on a ratio between the area of the segmentation mask and the area of the image to be annotated; The saliency prior graph fitness score is determined based on the correlation between the segmentation mask and the saliency prior graph corresponding to the image to be annotated.
7. The image labeling method according to claim 1, wherein the screening the plurality of segmentation masks based on the integrated saliency scores corresponding to the plurality of segmentation masks to determine the plurality of target masks corresponding to the image to be labeled comprises: Determining the segmentation mask with the highest comprehensive significance score as a basic optimal mask; Determining a plurality of original suboptimal masks except the basic optimal mask in a plurality of segmentation masks based on the basic optimal mask and the comprehensive significance scores corresponding to the plurality of segmentation masks; performing overlapping degree analysis on each original sub-optimal mask and the basic optimal mask, determining the overlapping degree of the sub-optimal mask corresponding to each original sub-optimal mask, and screening the original sub-optimal mask with the overlapping degree of the sub-optimal mask smaller than a fourth preset overlapping degree threshold value as a target sub-optimal mask; And determining the basic optimal mask and the target suboptimal mask as target masks corresponding to the images to be annotated.
8. An image marking apparatus, comprising: the mask processing module is used for carrying out mask processing on the image to be marked, and determining a plurality of segmentation masks corresponding to the image to be marked and segmentation scores corresponding to each segmentation mask; The saliency prior analysis module is used for carrying out saliency prior analysis on a plurality of segmentation masks corresponding to the image to be marked and determining target prior scores corresponding to the plurality of segmentation masks; A comprehensive significance score determining module, configured to determine a comprehensive significance score corresponding to the plurality of segmentation masks based on the segmentation scores corresponding to the plurality of segmentation masks and the target prior score; the segmentation mask screening module screens a plurality of segmentation masks based on comprehensive significance scores corresponding to the segmentation masks, and determines a plurality of target masks corresponding to the image to be marked; And the image labeling module is used for labeling the image to be labeled based on a plurality of target masks corresponding to the image to be labeled.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the image annotation method according to any of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the image labeling method of any of claims 1-7.

Description

Image labeling method, device, computer equipment and storage medium Technical Field The present invention relates to the field of image processing technologies, and in particular, to an image labeling method, an image labeling device, a computer device, and a storage medium. Background The saliency target detection is one of basic tasks of computer vision, is used for simulating a visual attention mechanism of human beings, detects a foreground target with stronger saliency in an image, and is widely applied to the fields of image segmentation, target identification, image editing, visual compression, robot navigation and the like. Saliency target detection is usually achieved based on a deep learning model, and the deep learning model needs to be trained based on large-scale and high-quality image annotation data so as to accurately detect the saliency target of the image in the applied field. In the prior art, the image annotation data is usually realized through manual annotation, the manual annotation is usually low in annotation efficiency and high in annotation cost, and high-quality image annotation data is difficult to efficiently generate. Therefore, how to label images reliably with high efficiency and high quality so as to perform accurate model training is a technical problem to be solved currently. Disclosure of Invention The embodiment of the invention provides an image labeling method, an image labeling device, computer equipment and a storage medium, which are used for solving the technical problem of how to label images reliably with high efficiency and high quality. An image annotation method comprising: Carrying out mask processing on an image to be marked, and determining a plurality of segmentation masks corresponding to the image to be marked and segmentation scores corresponding to each segmentation mask; performing significance prior analysis on a plurality of segmentation masks corresponding to the image to be marked, and determining target prior scores corresponding to the plurality of segmentation masks; Determining a comprehensive significance score corresponding to the plurality of segmentation masks based on the segmentation scores corresponding to the plurality of segmentation masks and the target prior score; Screening a plurality of segmentation masks based on comprehensive significance scores corresponding to the segmentation masks, and determining a plurality of target masks corresponding to the image to be marked; and labeling the image to be labeled based on a plurality of target masks corresponding to the image to be labeled. An image annotation device comprising: the mask processing module is used for carrying out mask processing on the image to be marked, and determining a plurality of segmentation masks corresponding to the image to be marked and segmentation scores corresponding to each segmentation mask; The saliency prior analysis module is used for carrying out saliency prior analysis on a plurality of segmentation masks corresponding to the image to be marked and determining target prior scores corresponding to the plurality of segmentation masks; A comprehensive significance score determining module, configured to determine a comprehensive significance score corresponding to the plurality of segmentation masks based on the segmentation scores corresponding to the plurality of segmentation masks and the target prior score; the segmentation mask screening module screens a plurality of segmentation masks based on comprehensive significance scores corresponding to the segmentation masks, and determines a plurality of target masks corresponding to the image to be marked; And the image labeling module is used for labeling the image to be labeled based on a plurality of target masks corresponding to the image to be labeled. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the image annotation method described above when executing the computer program. A computer readable storage medium storing a computer program which when executed by a processor implements the image annotation method described above. According to the image labeling method, the device, the computer equipment and the storage medium, the segmentation mask and the segmentation score for representing the reliability of the segmentation mask are determined through mask processing of the image to be labeled, the target prior score for representing the significance of the segmentation mask is determined through significance prior analysis of the segmentation mask, the significance of the segmentation mask is comprehensively analyzed according to the segmentation score for representing the reliability of the segmentation mask and the target prior score for representing the significance of the segmentation mask, the comprehensive significance score capable of accurately and reliably representing the signi