A method and system for scene segmentation based on context priors

Abstract

The invention discloses a scene segmentation method and system based on context prior, and belongs to the field of pattern recognition. The method includes: constructing a scene segmentation network; the network includes a feature extraction module, a feature aggregation module, a context prior module and a feature fusion module; the first two modules sequentially perform feature extraction and feature aggregation on the input image; The obtained features are learned to obtain a context prior map, and the context prior map is learned to obtain intra-class prior and inter-class prior, which are weighted with the output of the feature aggregation module to obtain intra-class features and inter-class features correspondingly; feature fusion module, the feature map, intra-class features and inter-class features output by the feature extraction module are output after cascade fusion and up-sampling; input the scene image to be segmented into the trained scene segmentation model to obtain the segmentation result. The present invention can clearly capture intra-class features and inter-class features, effectively improving the accuracy of scene segmentation.

Description

technical field [0001] The present invention belongs to the technical field of pattern recognition, and more particularly, to a method and system for scene segmentation based on context prior. Background technique [0002] The goal of scene segmentation is to assign a class label to each pixel to provide comprehensive scene understanding. challenging question. [0003] Benefiting from the powerful feature representation of fully convolutional networks (FCNs), many methods achieve decent scene segmentation performance. However, due to the limitation of the convolutional layer structure, the context information provided by FCN is insufficient, resulting in an insufficient understanding of the scene and affecting the accuracy of the recognition results. Therefore, various methods have emerged to explore contextual information for more accurate segmentation results. At present, there are two main ways to aggregate context information: 1) Pyramid-based aggregation methods; thi...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a scene segmentation method and system based on context prior, the purpose of which is to solve the problem that the existing scene segmentation method has contextual confusion, resulting in low accuracy of scene segmentation results technical problem

Images