An image automatic marking method and a device based on depth learning

Abstract

The invention discloses an image automatic marking method and a device based on depth learning, the method comprises the following steps: extracting visual features of an image to be marked by using depth learning technology; constructing the candidate tag set of the image to be annotated by using the image library, and extracting the semantic features of the image to be annotated from the candidate tag set of the image to be annotated by using the depth learning technology, fusing visual features and semantic features of the image to be annotated to obtain high-level features of the image tobe annotated; according to the high-level features of the images to be annotated, calculating the probabilities of each label in the image library by depth learning technique, according to the high-level features of the image to be annotated, predicting the number of tags needed for the image to be annotated by using depth learning technology. According to the calculated label probability and thepredicted label number, the first N labels with the highest probability are used to label the label image to be labeled. The invention can establish the relationship between the low-level feature andthe high-level semantic tag, thereby improving the accuracy of the image labeling.

Description

technical field [0001] The invention belongs to the field of image processing, and more specifically, relates to a deep learning-based automatic image labeling method and device. Background technique [0002] An image is a portrait of an objective object, vividly describing the visual information of the object, and is one of the most important sources of information. Image annotation is to annotate the image with some rich and appropriate keywords that can accurately describe the content of the image. Due to the ability to describe images at the semantic level, image annotation has a wide range of applications not only in the field of image analysis and understanding, but also in urban management, biomedical engineering and other related disciplines. [0003] Traditional image annotation mainly uses manual methods to annotate images with several keywords. However, in the current era of big data, due to the shortcomings of time-consuming, labor-intensive, and strong subject...

AI Technical Summary

Problems solved by technology

However, there are two main problems in this automatic image labeling method: one is that the optimality of the predicted label cannot be guaranteed; the other is that it is difficult to use the generative model to capture the complex relationship between image features and labels

Uniformly labeling each image with the same number of labels will lead to the problem that some images with rich content are not fully represented, while some images with relatively simple content are marked with too many labels. Therefore, this unified labeling method Will affect the accuracy of the labeling model

Images