Face attribute recognition method and system based on parallel shared multi-task network

Abstract

The invention relates to a face attribute recognition method and system based on a parallel shared multi-task network. The method comprises the following steps: face attributes contained in a face attribute data set are divided into a partial attribute group and an overall attribute group; a parallel shared multi-task network is constructed, and the parallel shared multi-task network is composed of a shared sub-network and two specific task sub-networks, namely a partial attribute sub-network and an overall attribute sub-network; an attention mechanism is adopted in each specific task sub-network, so that the correlation between local and global features of the shared sub-network and the specific task sub-network is utilized; a loss function is set, and a self-adaptive penalty strategy is adopted to relieve the problem of class imbalance, so that the face attribute recognition rate is improved; the parallel shared multi-task network is trained through the face attribute data set; and a face image to be recognized is input into the trained parallel shared multi-task network model to realize face attribute recognition. The method and the system are beneficial to improving the accuracy of face attribute recognition.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a face attribute recognition method and system based on a parallel shared multi-task network. Background technique [0002] Facial attribute recognition refers to the use of computers to analyze and judge various attribute information of a human face contained in an image, such as age, gender, expression, face shape, etc. Facial attribute recognition contains rich and stable feature information possessed by human faces, and is a popular research topic in computer vision and pattern recognition. As the value and influence of artificial intelligence continue to expand, its face images are easily captured by cameras, cameras and other instruments, and face attribute recognition is also widely used in image generation, human-computer interaction, video surveillance, recommendation systems and other fields. [0003] The specific task of face attribute recognition ...

AI Technical Summary

Problems solved by technology

In this process, only the high-level abstract features from the end of each branch are used for the final attribute prediction, which means that the low-level shared information may disappear when reaching the high-level layers of the network, eventually resulting in low-level shared information containing valuable spatial information. Features and detailed facial features may not be fully explored to improve overall performance

[0008] 2) Because the existing technology is learned based on the entire face image, there is a lot of interference information from other irrelevant facial regions when predicting attributes

But for attribute recognition, there is no attention mechanism that can simultaneously consider global and local features from different network layers, as well as the relationship between shared layers and task-specific layers.

[0009] 3) There is a category imbalance problem in the existing face attribute datasets

For example, in the most commonly used CelebA data set, the "bald" attribute has very few positive samples and a very large number of negative samples. Such an imbalance will lead to overfitting of the model, thereby reducing the generalization ability of the model.

For multi-label data sets, it is very difficult to rebalance the data of multiple labels, because balancing one of the attributes will affect the balance of another attribute

Images