基于卷积神经网络的人脸关键点检测算法设计

Design of facial point detection algorithm based on convolutional neural network

人脸关键点检测是模式识别和计算机视觉等领域十分活跃的研究方向。为了解决人脸关键点检测问题并提高检测的准确度,提出了一种基于卷积神经网络的人脸关键点检测算法,设计了一个新的网络模型,针对神经网络可能出现的梯度弥散和梯度爆炸的问题进行了优化。新型网络主要分为两个部分,第一部分是对人脸的关键点的总体预测,确定出大概位置;第二部分是对各个关键点的细微调整,保证结果的准确。为了提高网络的鲁棒性,采用两个数据集进行训练,在较短的训练时间内得到一个比较可靠的网络模型。实验结果表明所提出的方法十分适合用做人脸关键点检测,准确率达96%。

Facial point detection is a direction in the fields of pattern recognition and computer vision. The work aims to solve the facial detection and improve the accuracy. A facial point detection algorithm based on convolutional neural network is proposed. A new network model was designed and optimized for the problems of gradient dispersion and gradient explosion. The network is mainly divided into two parts, the first part is to roughly predict the points, the second part is to adjust the key points. In order to improve the robustness of the network, two datasets are used for training. In this way, a more reliable model can be obtained in a shorter training time. The result shows that the method is very suitable for facial point detection, and the accuracy rate is 96%.