Hair segmentation method based on deep learning

Abstract

The invention discloses a hair segmentation method based on deep learning. The hair segmentation method comprises the steps that step S100 an original image is acquired; S200 face detection is performed on the original image so as to obtain face key points; S300 a face key part description graph is generated by using the face key points; S400 the original image and the face key part description graph are combined so as to acquire a four-channel image; S500 the four-channel image is inputted to a convolutional neural network model, and the probability that each pixel point is the hair is inferred through the convolutional neural network model so as to acquire a hair probability graph; and S600 hair segmentation is performed on the original image by using the hair probability graph. The defects of the conventional segmentation method can be effectively avoided so as to achieve the great hair segmentation effect without artificial intervention and realize automatic segmentation; the image of arbitrary resolution can be segmented; and memory occupation is small and operation speed is fast.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a hair segmentation method based on deep learning. Background technique [0002] Image segmentation has always been a relatively important branch in the field of computer vision. Nowadays, some interactive segmentation methods such as grabcut and other graph-based segmentation methods are the most widely used. These methods need to manually specify a part of the foreground and background seed points. To have a good user experience, automation is required as much as possible; graph cutting methods such as grabcut only use color information and spatial information for segmentation, and lack high-level semantic segmentation and understanding. [0003] Hair segmentation is a kind of situation in image segmentation. The complexity of hair style, the change of color, the uncertainty of texture, the uncertainty of brightness, etc. make the traditional image cutting method not a...

AI Technical Summary

Problems solved by technology

[0003] Hair segmentation is a kind of situation in image segmentation. The complexity of hair style, the change of color, the uncertainty of texture, the uncertainty of brightness, etc. make the traditional graph cutting method not able to segment hair well.

However, the existing hair segmentation method makes the effect of hair segmentation poor, and it is impossible to clearly segment the hair area; it cannot realize automatic segmentation and requires manual intervention; it cannot perform hair segmentation on images of arbitrary resolution; it takes up a lot of memory and runs slowly

Images