A human body parsing and segmentation model and method based on edge information enhancement

Abstract

The invention discloses a human body parsing and segmentation model and method based on edge information enhancement, and belongs to the technical field of semantic segmentation in computer vision. The network structure of the human body analysis segmentation model of the present invention is based on ResNet-101, and an edge feature extraction module and a human body feature extraction module are constructed, which are used to extract edge feature maps and human body analysis feature maps respectively. The edge feature extraction module and the human body feature extraction module take the four features generated by ResNet‑101 as input to further extract and enhance features. Both modules use DenseASPP to extract multi-dimensional features. On this basis, the network structure is further planned to extract effective features corresponding to the two tasks. The model loss item only includes human body segmentation loss and edge loss, which is easy to adjust the weight in the training process, and can more specifically explore the potential of edge information. The invention is used for multi-category fine segmentation of a single human body, and its segmentation performance is better than that of the existing segmentation methods.

Description

A Human Analytical Segmentation Model and Method Based on Edge Information Enhancement technical field The invention belongs to the semantic segmentation field in computer vision, be specifically related to a kind of utilizing edge information to enhance human body Parsing segmentation techniques. Background technique [0002] Human parsing is a subtask of semantic segmentation. Its goal is to combine the various Parts or clothing accessories to be identified. Unlike general semantic segmentation, human parsing focuses on human-centric segmentation. To cut, it is necessary to identify the arms, head, legs and other areas of the human body, that is, to make detailed segmentation of each part of the human body. Human Parsing in Behavior Recognition, pedestrian re-identification, fashion synthesis and other fields have applications. [0003] Research techniques try to use cutting-edge deep learning techniques to improve model performance, such as: based on multi-task l...

AI Technical Summary

Problems solved by technology

[0006] The problems that need to be solved in scene analysis are more complex than human body analysis, but as far as the existing mainstream data sets are concerned, the data sets such as LIP and CIHP in the field of human body analysis contain 20 categories, and the commonly used data set in the field of scene analysis Cityscape Contains 19 categories, which means that the complexity of the human body analysis model is comparable to that of scene analysis, so it is necessary to set up the human body analysis model by further mining edge information to enrich the feature dimensions extracted by the human body analysis model and achieve better performance. Segmentation performance

Images