A Face Depth Prediction Method Against Grid Effect

Abstract

Provide a face depth prediction method that is resistant to grid effects, including steps: S1: Build a convolutional neural network that is resistant to grid effects, including multiple hole convolutions, normalization operations and excitation operations connected in series for each hole convolution , and multiple pixel deconvolution; S2: Establish a face data set, which includes a training set and a test set, and set the training parameters of the convolutional neural network; S3: Initialize the weight of the convolutional neural network, and input the training set In the convolutional neural network, the neural network is trained with the goal of minimizing the cost function to form a facial image depth prediction neural network model; S4: Input the test set into the facial image depth prediction neural network model, and the output can reflect the depth information of the facial image Image. The face depth prediction method of the present invention can solve the grid effect of the traditional convolutional neural network, has a larger receptive field, can greatly improve the accuracy of face depth prediction, and is beneficial to the three-dimensional reconstruction research of the face.

Description

technical field [0001] The invention relates to the field of computer vision and digital image processing, in particular to a face depth prediction method resistant to grid effects Background technique [0002] Facial depth prediction is a novel and challenging problem in computer vision. Depth prediction for faces is an important part of understanding the geometry of the whole face. Correspondingly, such a relationship can well reflect the organs on the human face and the environment of the entire human face. If better facial depth information can be obtained, it will be of great help to the face recognition problem. , and also help to build a 3D model of the human face, which is beneficial to solve the problem of 3D reconstruction of the human face. However, predicting the depth information of the entire face from an RGB image of a face is a problem with a pathological nature, because there are many uncertainties in the process of mapping the color information of the RGB...

AI Technical Summary

Problems solved by technology

However, there is a common grid problem in hole convolution, because although hole convolution can expand the receptive field, it fills the convolution kernel with 0 to form a grid effect. For hole convolution with a hole rate of 2, it is large Dating loses 75% of information, and as the hole rate continues to increase, local information will be lost

The same situation also exists in the deconvolution of the upsampling operation. In the process of upsampling, there is no direct connection between adjacent pixels, resulting in a grid effect, which affects the accuracy of the generated depth map.

Images