Image processing method and apparatus, computer-readable storage medium

1. An image processing method, comprising:

obtaining a to-be-processed image;

performing image semantic segmentation on the to-be-processed image to obtain a semantically-segmented image, the semantically-segmented image comprising a target region and a non-target region obtained through the semantic segmentation;

inputting the to-be-processed image into a pose recognition model;

partitioning an image region in which the target in the to-be-processed image is located using a first hidden layer of the pose recognition model;

determining first skeletal key points corresponding to the target in the image region using a second hidden layer of the pose recognition model, the first hidden layer being located in front of the second hidden layer;

determining a plurality of skeletal key points in the to-be-processed image using the first hidden layer of the pose recognition model;

clustering the plurality of skeletal key points according to targets in the to-be-processed image using the second hidden layer of the pose recognition model, to obtain second skeletal key points corresponding to the target;

outputting, using the pose recognition model, a pose-recognized image recognizing skeletal region, the skeletal region being predicted according to the first skeletal key points and the second skeletal key points;

fusing the target region and the non-target region of the semantically-segmented image with the skeletal region of the pose-recognized image, to obtain a trimap comprising foreground region, background region, and recognition region; and

generating, according to the to-be-processed image and the trimap, a transparency mask image for image separation from the to-be-processed image.

2. The method of claim 1 , wherein the performing image semantic segmentation on the to-be-processed image to obtain the semantically-segmented image comprises:

inputting the to-be-processed image into an image semantic segmentation model;

outputting a semantic category probability distribution matrix using the image semantic segmentation model; and

generating the semantically-segmented image according to the semantic category probability distribution matrix, wherein pixels in the semantically-segmented image have pixel values representing semantic categories to which the pixels belong and are corresponding to pixels in the to-be-processed image, pixels in the target region belong to a target semantic category, and pixels in the non-target region do not belong to the target semantic category.

3. The method of claim 1 , wherein the fusing the target region and the non-target region of the semantically-segmented image with the skeletal region of the pose-recognized image to obtain the trimap comprises:

expanding a region boundary between the target region and the non-target region in the semantically-segmented image; and

using a remaining non-target region in the semantically-segmented image after the expansion as the background region and using a region corresponding to the skeletal region of the pose-recognized image in the semantically-segmented image as the foreground region, to obtain the trimap comprising the foreground region, the background region, and recognition region, the recognition region of the trimap being a remaining region in the semantically-segmented image excluding the background region and the foreground region.

4. The method of claim 3 , wherein the expanding the region boundary between the target region and the non-target region in the semantically-segmented image comprises:

performing edge detection on the target region and the non-target region to obtain the region boundary between the target region and the non-target region; and

expanding the region boundary in the semantically-segmented image toward two sides using the region boundary as a center.

5. The method of claim 1 , wherein the generating the transparency mask image for separating image from the to-be-processed image comprises:

performing global sampling on target pixels in a recognition region of the to-be-processed image according to the trimap, to select matching foreground pixels and matching background pixels for the target pixels;

fusing the matching foreground pixels and the matching background pixels, to obtain transparency corresponding to the target pixels; and

generating, according to the transparency corresponding to the target pixels, the transparency mask image for image separation from the to-be-processed image.

6. The method of claim 5 , wherein the generating the transparency mask image for image separation from the to-be-processed image comprises:

generating an intermediate image by using the transparency corresponding to the pixels in the trimap as pixel values of the pixels; and

generating, after filtering and image morphology operations are performed on the intermediate image, the transparency mask image for image separation on the to-be-processed image.

7. The method of claim 1 , wherein the target region is a human region, the non-target region is a non-human region, and the performing the pose recognition on the to-be-processed image, to obtain the pose-recognized image comprises:

performing human pose recognition on the to-be-processed image, to obtain a pose-recognized image recognizing human skeletal region; and

the generating the transparency mask image for image separation from the to-be-processed image comprises:

generating, according to the to-be-processed image and the trimap, the transparency mask image for performing human image matting on the to-be-processed image.

8. The method of claim 1 , further comprising:

performing an operation on pixel values of pixels in the to-be-processed image and pixel values of corresponding pixels in the transparency mask image; and

separating a foreground region from the to-be-processed image according to a result of the operation.

9. The method of claim 8 , wherein the obtaining the to-be-processed image comprises:

obtaining a to-be-processed video;

using a video frame in the to-be-processed video as the to-be-processed image; and

the method further comprises:

performing an editing operation on the to-be-processed image from which the foreground region has been separated, to obtain a corresponding edited image; and

generating a target video according to the edited image.

10. An image processing apparatus, comprising:

a memory operable to store computer-readable instructions; and

a processor operable to read the computer-readable instructions, the processor when executing the computer-readable instructions is configured to:

obtain a to-be-processed image;

perform image semantic segmentation on the to-be-processed image to obtain a semantically-segmented image, the semantically-segmented image comprising a target region and a non-target region obtained through the semantic segmentation;

input the to-be-processed image into a pose recognition model;

partition an image region in which the target in the to-be-processed image is located using a first hidden layer of the pose recognition model;

determine first skeletal key points corresponding to the target in the image region using a second hidden layer of the pose recognition model, the first hidden layer being located in front of the second hidden layer;

determine a plurality of skeletal key points in the to-be-processed image using the first hidden layer of the pose recognition model;

cluster the plurality of skeletal key points according to targets in the to-be-processed image using the second hidden layer of the pose recognition model, to obtain second skeletal key points corresponding to the target;

output, using the pose recognition model, a pose-recognized image recognizing skeletal region, the skeletal region being predicted according to the first skeletal key points and the second skeletal key points;

fuse the target region and the non-target region of the semantically-segmented image with the skeletal region of the pose-recognized image, to obtain a trimap comprising foreground region, background region, and recognition region; and

generate, according to the to-be-processed image and the trimap, a transparency mask image for image separation from the to-be-processed image.

11. The apparatus of claim 10 , wherein the processor is configured to:

input the to-be-processed image into an image semantic segmentation model;

output a semantic category probability distribution matrix using the image semantic segmentation model; and

generate the semantically-segmented image according to the semantic category probability distribution matrix, wherein pixels in the semantically-segmented image have pixel values representing semantic categories to which the pixels belong and are corresponding to pixels in the to-be-processed image, pixels in the target region belong to a target semantic category, and pixels in the non-target region do not belong to the target semantic category.

12. The apparatus of claim 10 , wherein the processor is configured to:

expand a region boundary between the target region and the non-target region in the semantically-segmented image; and

use a remaining non-target region in the semantically-segmented image after the expansion as the background region and using a region corresponding to the skeletal region of the pose-recognized image in the semantically-segmented image as the foreground region, to obtain the trimap comprising the foreground region, the background region, and recognition region, the recognition region of the trimap being a remaining region in the semantically-segmented image excluding the background region and the foreground region.

13. The apparatus of claim 10 , wherein the processor is configured to:

perform global sampling on target pixels in a recognition region of the to-be-processed image according to the trimap, to select matching foreground pixels and matching background pixels for the target pixels;

fuse the matching foreground pixels and the matching background pixels, to obtain transparency corresponding to the target pixels; and

generate, according to the transparency corresponding to the target pixels, the transparency mask image for image separation from the to-be-processed image.

14. A non-transitory computer-readable storage medium, having processor executable instructions stored thereon for causing a processor to:

obtain a to-be-processed image;

perform image semantic segmentation on the to-be-processed image to obtain a semantically-segmented image, the semantically-segmented image comprising a target region and a non-target region obtained through the semantic segmentation;

input the to-be-processed image into a pose recognition model;

partition an image region in which the target in the to-be-processed image is located using a first hidden layer of the pose recognition model;

determine first skeletal key points corresponding to the target in the image region using a second hidden layer of the pose recognition model, the first hidden layer being located in front of the second hidden layer;

determine a plurality of skeletal key points in the to-be-processed image using the first hidden layer of the pose recognition model;

cluster the plurality of skeletal key points according to targets in the to-be-processed image using the second hidden layer of the pose recognition model, to obtain second skeletal key points corresponding to the target;

output, using the pose recognition model, a pose-recognized image recognizing skeletal region, the skeletal region being predicted according to the first skeletal key points and the second skeletal key point;

fuse the target region and the non-target region of the semantically-segmented image with the skeletal region of the pose-recognized image, to obtain a trimap comprising foreground region, background region, and recognition region; and

generate, according to the to-be-processed image and the trimap, a transparency mask image for image separation from the to-be-processed image.