Few-sample document layout analysis method based on metric learning

Abstract

The invention discloses a few-sample document layout analysis method based on metric learning, and the method comprises the steps: constructing position attention features and channel attention features of different regions through extracted original feature graphs of a document image, and enabling the representation features of different regions in the feature graphs to be fully utilized throughthe fusion of a convolution network; carrying out the prototype construction by using representation features obtained by fusion, and carrying out the alignment operation by using a result obtained bysegmentation, so document content segmentation is carried out by using a small amount of data, and meanwhile, the segmentation precision is improved.

Description

technical field [0001] The invention belongs to the technical field of semantic segmentation in computer vision, and more specifically, relates to a method for analyzing the layout of a few-sample document based on metric learning. Background technique [0002] In the era of mobile Internet, the acquisition and sharing of electronic documents has become very convenient, and the analysis of document layout can effectively extract valuable information. As the number of documents continues to increase and the contents of documents become more diverse, analyzing the layout content of documents has become a new trend of semantic segmentation. The goal of document layout analysis is to classify different regions in the document image and obtain segmentation results with different label information. [0003] The existing method for analyzing document layout is better based on semantic segmentation based on deep network. The semantic segmentation method based on deep network mainly...

AI Technical Summary

Problems solved by technology

[0006] Although based on the above two methods and derived many new deep semantic segmentation networks with remarkable effects, these networks all have the same serious problem: these networks require a large amount of finely labeled data for training

However, the training data requires pixel-by-pixel marking, which makes the acquisition of training data a very time-consuming and laborious task, especially in the case of complex document content.

One way to delay is to use weakly supervised learning for training, but still requires a lot of weakly labeled training data

Images