Face analysis device and method

Abstract

The invention provides a face analysis device and method. The face analysis method comprises that a to-be-detected sample is input to a residual error network module; the trained residual error network module is used to process the to-be-detected sample, the residual error network module comprises residual error blocks which are arranged from the input to the output direction and combined sequentially, and output of the Nth predetermined residual error block among the residual blocks combined sequentially is sent to a residual error deconvolution network module; and the trained residual errordeconvolution network module is used to process the output of the Nth residual error block to obtain a classifcaitn graph, the residual error deconvolution network module comprises residual error deconvolution blocks combined sequentially, and the residual error deconvolution blocks correspond to the first to Nth residual error blocks respectively. Via the face analysis method, the face analysis performance can be improved, and the model size in the method is reduced greatly.

Description

technical field [0001] The invention relates to an image processing device and an image processing method related to computer vision, in particular to a face analysis device and a face analysis method including a residual deconvolution network. Background technique [0002] Computer vision refers to the use of cameras and computers instead of human eyes to perform machine vision operations such as recognition, tracking and measurement of targets, and further graphics processing so that the signal becomes an image more suitable for human observation or instrument detection. Computer vision uses computers and related equipment to simulate biological vision. The ultimate research goal is to enable computers to observe and understand the world through vision like humans, and have the ability to adapt to the environment autonomously. In today's intelligent and digitalized world, computer vision has been widely used and paid great attention to. [0003] The use of deep learning f...

AI Technical Summary

Problems solved by technology

First, as the number of network layers deepens, it will cause gradient dispersion or network training decline, making it difficult for the entire network to reach a better position

Second, the stacked deconvolution network always uses the rough feature map of the previous layer as input, and expects to output a fine feature map through deconvolution. The solution space is large and it is not easy to converge

However, the current existing technology does not have a good solution for how to use prior information

Images