Image segmentation method and device and storage medium

Abstract

The embodiment of the invention provides an image segmentation method and device, and a storage medium. The image segmentation method comprises the steps: carrying out the pyramid down-sampling of a sample image through an initial classification network, and obtaining the multilayer sample image features corresponding to the image; converting a plurality of image channel values corresponding to aplurality of resolutions of the multi-layer sample image features into first image channel values, and converting a plurality of resolutions of the multi-layer sample image features into first resolutions to obtain a plurality of sample feature maps, wherein the image channel values and the resolutions of the plurality of sample feature maps are the same; obtaining a target segmentation sample image by using the plurality of sample feature maps; training the initial classification network by using the target segmentation sample image and a standard segmentation sample image corresponding to the sample image to obtain a preset classification network; and inputting the to-be-segmented image into a preset classification network to obtain a target segmented image corresponding to the to-be-segmented image.

Description

technical field [0001] The present application relates to the field of image processing, in particular to an image segmentation method and device, and a storage medium. Background technique [0002] Image segmentation is a basic topic in the field of computational vision, and portrait segmentation is one of the most important applications. In applications such as portrait blurring and background replacement using smart terminals, high-precision portrait segmentation technology is required. With the rapid development of deep learning, convolutional neural networks are usually used to handle portrait segmentation tasks. Specifically, in the encoding stage, the image passes through a certain number of sequentially connected convolution-downsampling layers, and outputs the downsampled feature map in the encoding stage; in the decoding stage, the downsampled feature map passes through a certain number of sequentially connected convolution-upsampling layers. Layers generate upsa...

AI Technical Summary

Problems solved by technology

[0003] However, the existing deep learning models for portrait segmentation often only use the feature map with the lowest resolution and the highest number of channels when reusing the features extracted by the basic network, or the use of feature maps is very limited, thereby reducing the image quality. segmentation accuracy

Images