An image classification method and device

Abstract

The embodiment of the present application discloses an image classification method and device. For the image to be processed, N labeling results of the image to be processed can be respectively determined through N labeling models, and the label consistency can be determined comprehensively from the overall perspective according to the N labeling results. Consistency parameters, reduce the impact of labeling deviation in a certain model on image classification, and improve the credibility of labeling consistency parameters. The determined labeling consistency parameter can reflect the labeling consistency degree of the N labeling models for the image to be processed, so as to determine the labeling difficulty of the image to be processed, and determine the image to be processed as the first type of image that is easy to label, or is not easy to label The second type of image. By classifying the difficulty level of labeling for the images to be processed, the first type of images that are easy to label and the second type of images that are not easy to label can be distinguished, so that the labeling strategy can be arranged according to the difficulty of labeling, and the cost of labeling can be reduced. Improve labeling efficiency.

Description

technical field [0001] The present application relates to the field of data processing, and in particular, to an image classification method and apparatus. Background technique [0002] Image labeling refers to labeling the content contained in an image to identify the content, such as identifying the type, name, shape, and appearance of the content. [0003] In some fields, images are mainly annotated manually. For example, in the medical field, doctors make corresponding annotations after analyzing medical images. [0004] The manual annotation method relies too much on the human experience of the annotator, which not only has low labeling efficiency, but also the labeling accuracy and labeling cost of images by different experienced labelers are often very different. Although experienced labelers can have better labeling accuracy However, the annotation cost is much higher than that of experienced annotators. SUMMARY OF THE INVENTION [0005] In order to solve the abo...

AI Technical Summary

Problems solved by technology

[0004] The way of manual labeling relies too much on the human experience of the labeler, not only the labeling efficiency is low, but also the labeling accuracy and labeling cost of images by different experienced labelers are often very different, although experienced labelers can have better labeling accuracy , but the labeling cost is much higher than that of experienced labelers

Images