基于跳连接反卷积神经网络的自动眼镜摘除

Automatic glass removal via skip-connected deconvolutional neural networks

随着人工智能的快速发展,人脸识别已经成为计算机视觉领域非常热门的研究方向;眼镜作为最常见的遮挡物,对人脸识别的性能有着较大的影响。传统方法大多使用主成分分析(principal component analysis,PCA)重建技术对眼镜进行摘除,但是预先定位眼镜位置非常困难,而且重建后图像的眼镜区域和非眼镜区域有着明显的不连续性。针对眼镜遮挡这一问题,文章提出了一种基于跳连接反卷积神经网络的自动眼镜摘除方法,实现了端对端的自动眼镜摘除技术,在尽可能保留人脸细节的前提下将眼镜从人脸图片中去除。该方法较好地解决了PCA重建中眼镜区域和非眼镜区域不连续性问题,使得眼镜去除效果更加接近真实效果。

With the rapid development of artificial intelligence,face recognition has become a very popular research direction in the field of computer vision.Glasses,as one of the most common obstructions,tend to have great influence on the performance of face recognition.For most of traditional methods,the principal component analysis(PCA)is used to reconstruct the glassless face,but the PCA need to detect glass areas before reconstruction which is a great challenge and there will be obvious discontinuity between the boundary of occluded and disoccluded areas.In view of the above mentioned problem,an algorithm of automatic glass removal via skip-connected deconvolutional neural networks is proposed.The algorithm can achieve the end-to-end automatic glass removal,removing glasses from face images while keeping more facial details.The algorithm solves the problem of discontinuity between the boundary of occluded and disoccluded areas in PCA method and makes the effect of glass removal closer to the real one.