Method and device for extracting human head image based on deep learning

Abstract

The invention discloses a method and device for extracting a human head image based on deep learning. The method comprises the steps of inputting a to-be-extracted picture including a portrait into apre-trained YOLO neural network for processing and outputting a head region image, inputting the head region image into a trained HED neural network for processing and outputting a head edge contour image, inputting the head region image serving as a content image and the head edge contour image serving as a style image into an artistic style transfer neural network for processing, and thus outputting a synthetic image. The device comprises a memory and a processor. The head edge contour image can be extracted from the head region image, and the style of the head edge contour image is transferred to the head region image, thereby realizing the integration of the details such as the head edge contour, the five sense organs and the hairs, and enabling the synthetic image to have rich facialfeatures and clear edge contour. The method and device disclosed by the invention are applied to the technical field of image recognition and processing.

Description

technical field [0001] The present invention relates to the technical field of image recognition processing, in particular to a method and device for extracting human head images based on deep learning. Background technique [0002] Explanation of terms: [0003] YOLO: You Only Look Once, is a target detection method based on deep learning, which can be used to train neural networks. The neural network trained by YOLO can be used to solve the regression problem of target area prediction and category prediction. Its advantage is that it can At the same time, it ensures high detection speed and accuracy; [0004] HED: Holistically-nested Edge Detection, which is embedded in edge detection as a whole, is a deep neural network model. When it is used for image edge contour detection, the input object is a picture, and the output object is the edge contour image of the main shape in the picture. [0005] With the development of robot technology, robots are more and more widely u...

AI Technical Summary

Problems solved by technology

These existing detection methods require human participation to adjust the detection results, and pay too much attention to the facial details of the head image, which may easily cause the edge of the face in the recognition result to break, which affects the portrait effect of the portrait robot

Images