Image semantic annotation method and device and storage medium

Abstract

The invention provides an image semantic annotation method and device and a storage medium. The method comprises the steps: carrying out the layering of a to-be-annotated image, and obtaining N layersof to-be-annotated sub-images corresponding to the to-be-annotated image; obtaining a first probability that each pixel block in each layer of sub-to-be-annotated image belongs to a preset object corresponding to each layer of sub-to-be-annotated image according to the N layers of sub-to-be-annotated images and a corresponding relationship between preset pixel blocks and the probability that thepreset pixel blocks belong to each preset object; according to the plurality of first probabilities corresponding to each layer of sub-images to be labeled, obtaining a ternary graph corresponding toeach layer of sub-images to be labeled; and labeling the ternary graph corresponding to each layer of sub-to-be-labeled image to obtain a labeling result of the to-be-labeled image. According to the invention, the annotation result of the to-be-annotated image is obtained according to the probability that each pixel block in the multiple layers of sub-to-be-annotated images belongs to the preset object, so that the annotation result in the to-be-annotated image is more accurate, and the problem of manual modification is avoided.

Description

technical field [0001] The present invention relates to the technical field of image semantic annotation, in particular to an image semantic annotation method, device and storage medium. Background technique [0002] Image semantic annotation can be said to be the cornerstone technology of image understanding, and plays a pivotal role in autonomous driving, drone applications, and wearable device applications. An image is composed of multiple pixel blocks (Pixel), and semantic annotation, as the name implies, is to identify and label pixel blocks according to the different semantic meanings expressed in the image. [0003] In the prior art, public image annotation data is usually used to train a deep neural network model, and then the model is used to predict the annotation of a new image, and manually modify and adjust the annotated image to generate a semantic annotation result of the input image. [0004] However, the labeling results output by the deep neural network mo...

AI Technical Summary

Problems solved by technology

[0004] However, the labeling results output by the deep neural network model in the prior art need to be further modified manually, and there are a large number of target (contour) boundaries in the image to be labeled, which are limited by the image resolution and the resolution of the display, and the boundaries are often blurred , resulting in a large amount of manual modification and a high error rate

Images