基于卷积神经网络的 U-net 模型的 人像分离工具 portrait separation tool of u-net model based on convolutional neural network内容摘要图像信息是人类视觉感知信息的重要部分,随着计算机、电子技术日益进步,大规模运算和信息的数-模转换走入现实,由此催生出分为图像识别、分割、重建等领域的数字处理技术蓬勃发展。作为图像分割的典型应用场景,人像分割是进行人像美化、背景处理、人脸识别等领域的重要前提。因此通过人像分割从图像中精确且完整地提取出人像目标,对后续处理有必要作用。本文以深度学习中的卷积神经网络为基础,研究如何利用 U-Net 模型在人像照片较有限的情况下达到人像分离合格的准确率。最终实现模型的 IOU 在 92.1%左右,Dice 系数在 95.7%左右。对比早期的分割方法,卷积神经网络 U-Net 模型通过类似编码器—解码器的“U”型结构实现更简便、稳定地提取数据的特征,且能更加有效的提高分割的准确度。关键词:深度学习;人像分割;U-Net 模型AbstractImage information is an important part of human visual perception information. With the development of computer and electronic technology, large-scale operation and digital-to-analog conversion of information come into reality, Therefore, the digital processing technology including image recognition, segmentation and reconstruction has been developed vigorously.. As a typical application scene of image segmentation, is an important premise of image beautification, background processing, face recognition and so on. Therefore, it is necessary to extract the image object accurately and completely from the image by portrait segmentationIn this paper, based on the convolutional neural network in deep learning, we study how to use the U-Net model to achieve the qualified accuracy rate of human image separation under the limited situation of human image. The IOU of the final model is about 92.1% , and the Dice coefficient is about 95.7% . Compared with the earlier segmentation methods,The convolutional...