Image automatic annotation method and system based on attention disturbance and medium

Abstract

The invention discloses an automatic image annotation method and system based on attention disturbance and a medium, and the method comprises the following steps: dividing a to-be-annotated data set into a labeled data set and a non-labeled data set, carrying out the image enhancement of the labeled data set to expand the data set, and carrying out the annotation through a corresponding label; constructing a generative adversarial network based on an attention disturbance mechanism, wherein the generative adversarial network comprises an image auto-encoder, a generator and a discriminator; iteratively training a generative adversarial network by using a to-be-labeled data set, and optimizing a multi-task full-connection classification network of a discriminator by using an expanded labeled data set; and using the trained discriminator to carry out classification labeling on the to-be-labeled images without labels. According to the method, the generative adversarial network based on the attention disturbance mechanism is constructed, meanwhile, the multi-task full-connection classification network is introduced, image feature representation is optimized through adversarial training, automatic annotation of the image is achieved, and diversity and quality of image generation are improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an attention perturbation-based automatic image labeling method, system and medium. Background technique [0002] As an important way of carrying information in modern society, images are widely used in various industries due to their natural fit with human visual perception. However, in today's big data era of information explosion, it is precisely because of the wide application of images that the number and categories of image data have increased significantly, and it has become more difficult to screen and filter specific categories of images. For example, social platforms need to retrieve and delete inappropriate content from the massive images uploaded every day, and manually reviewing content is not only too much work, but also inefficient. [0003] In recent years, with the emergence of artificial intelligence technology, complex tasks such as image classificati...

AI Technical Summary

Problems solved by technology

Although the generative confrontation network has a strong learning ability, if there is a lack of supervision information during training, problems such as mode collapse, catastrophic forgetting, and cyclic performance will occur, which will prevent the network from learning effective representations of the characteristics of the data set.

Images