Portrait segmentation method based on cascaded convolutional neural network

Abstract

The invention discloses a portrait segmentation method based on a cascaded convolutional neural network. The method specifically comprises the following steps: (1) searching a large amount of portraitdata, and manually labeling the data to obtain a binary portrait labeling mask consistent with an original image; (2) constructing multi-scale image input; (3) constructing a primary portrait segmentation network; (4) constructing a secondary portrait segmentation network; (5) constructing input of a secondary portrait segmentation network; (6) constructing a loss function of the whole network; (7) performing back propagation on the whole network according to the loss function of the whole network to update the weight, and obtaining a trained portrait segmentation model. The beneficial effects of the invention are that the method achieves the better restoration of the portrait edge under the condition that the complexity of the model is slightly increased; the portrait integrity is maintained while the portrait edge is optimized by the secondary network; and the overall portrait segmentation effect and the portrait edge accuracy are improved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a method for segmenting portraits based on a cascaded convolutional neural network. Background technique [0002] Portrait segmentation refers to separating the portrait from the background in the picture. The separated portrait will provide the basis for subsequent applications such as background blur, background replacement, and portrait movement. Due to the high complexity of the background and the diversity of portrait postures, traditional portrait segmentation cannot extract portrait areas well. Therefore, most current portrait segmentation technologies are based on deep learning methods. [0003] Portrait segmentation methods based on deep learning usually use fully convolutional neural networks to segment portraits and backgrounds in an end-to-end predictive manner, such as network architectures such as FCN, U-net, and DeepLab. However, these methods gen...

AI Technical Summary

Problems solved by technology

However, these methods generally have the problem of inaccurate edge segmentation, or increase the accuracy of the results by increasing the size of the network model, resulting in increased algorithm complexity and slower running speed.

Therefore, it is impossible to achieve a good balance in terms of effect and performance, and it is difficult to meet the current needs

Images