当前成熟的人脸识别技术已广泛应用于多个领域。然而，在一些难以获取人脸照片的特殊场景下，通常采用通过知情者的面部特征回忆绘制人脸的草图，并寻找目标人物。在该场景下，如何通过人脸草图准确识别人物身份这一问题至关重要。为解决以上问题，本文设计了一种面向人脸草图分析与识别的方法。以深度神经网络结构为基础，本文提出将传统的人脸面部特征算法与卷积神经网络特征提取算法相结合，将SIFT特征描述子与高维抽象的特征进行结合的提取与表示，并通过传统人脸区域特征的检测和识别，调整不同区域图像特征的权重以实现算法的注意力机制。并且本文采用孪生网络架构将图像和草图进行特征比对，通过三元损失函数优化训练得到的特征表示来计算图像特征的距离，最终获得识别的结果。本文所设计的方法能够更准确地处理人脸草图识别中特征不平衡与带有遮挡情况的问题。 Face recognition has attracted wide attention. However, in some special scenes, it is difficult to directly obtain the facial photo shoot of the target person, so the facial features of witnesses are used to recall, draw a hand-drawn sketch of the human face, and use this sketch to find the target person. In this scenario, it is crucial to accurately identify people through face sketches. There is an increasing need for more accurate and reliable sketching and face authentication technologies. To address these issues, we devised a method for face sketch analysis and recognition. The presented algorithm is based on the deep neural network combined with the traditional image feature extraction algorithm expressed attention mechanisms. Then it uses Siamese network to compare sketch image features. The designed method can handle feature imbalance and shielding problem more accurately in face sketch recognition.
展开▼
机译：当前成熟的人脸识别技术已广泛应用于多个领域。然而，在一些难以获取人脸照片的特殊场景下，通常采用通过知情者的面部特征回忆绘制人脸的草图，并寻找目标人物。在该场景下，如何通过人脸草图准确识别人物身份这一问题至关重要。为解决以上问题，本文设计了一种面向人脸草图分析与识别的方法。以深度神经网络结构为基础，本文提出将传统的人脸面部特征算法与卷积神经网络特征提取算法相结合，将SIFT特征描述子与高维抽象的特征进行结合的提取与表示，并通过传统人脸区域特征的检测和识别，调整不同区域图像特征的权重以实现算法的注意力机制。并且本文采用孪生网络架构将图像和草图进行特征比对，通过三元损失函数优化训练得到的特征表示来计算图像特征的距离，最终获得识别的结果。本文所设计的方法能够更准确地处理人脸草图识别中特征不平衡与带有遮挡情况的问题。 Face recognition has attracted wide attention. However, in some special scenes, it is difficult to directly obtain the facial photo shoot of the target person, so the facial features of witnesses are used to recall, draw a hand-drawn sketch of the human face, and use this sketch to find the target person. In this scenario, it is crucial to accurately identify people through face sketches. There is an increasing need for more accurate and reliable sketching and face authentication technologies. To address these issues, we devised a method for face sketch analysis and recognition. The presented algorithm is based on the deep neural network combined with the traditional image feature extraction algorithm expressed attention mechanisms. Then it uses Siamese network to compare sketch image features. The designed method can handle feature imbalance and shielding problem more accurately in face sketch recognition.
展开▼
