Face image data clustering method, system and device

Abstract

The invention discloses a face image data clustering method, which comprises the steps of inputting a face image into a preset feature extraction model to extract feature data; calculating a similarity score between the feature data of any two face images by adopting a cosine similarity function; acquiring two face images of which the similarity scores are greater than a preset first similarity threshold as candidate image pairs in the image pair set; calculating feature vectors of the candidate image pairs, and inputting the feature vectors into a preset classifier to enable the classifier toperform recognition; when the classifier judges that the candidate image pairs do not belong to the same person, removing the candidate image pairs from the image pair set; and merging and classifying all candidate image pairs belonging to the same person in the image pair set into one class. The invention further discloses a face image data clustering system and a face image data clustering device. By adopting the embodiment of the invention, the clustering deviation can be effectively reduced according to the distribution characteristics of the human face samples collected in the real scene.

Description

technical field [0001] The invention relates to the technical field of face recognition, in particular to a face image data clustering method, system and equipment. Background technique [0002] With the rapid development of deep learning, face recognition algorithms based on deep neural networks have been widely used in various fields. In order to obtain a more effective and stable deep learning face recognition model, it is usually necessary to use a large number of face samples for training and learning, and the labor and time costs of labeling and cleaning these training samples are very high. Therefore, some existing technologies propose the use of clustering algorithms to automatically label a large number of unlabeled face images. The most commonly used clustering algorithms are K-means clustering, spectral clustering, hierarchical clustering and so on. [0003] In the process of using the K-means clustering algorithm, it is calculated by assuming that the samples i...

AI Technical Summary

Problems solved by technology

[0003] In the process of using the K-means clustering algorithm, it is calculated by assuming that the samples in a certain class will be tightly distributed around a cluster center. However, the face images of the same person in the real scene will vary due to the angle, Factors such as lighting lead to a more complex distribution in the actual distribution, so the clustering algorithm can easily aggregate a person's face image into multiple categories

The spectral clustering algorithm requires that the number of samples of different categories in the data set is balanced, that is, the number of face images of different people is not much different, but the real collected data often cannot meet this assumption, and some people may only collect a single image. Zhang, and some people's images will be very many

Therefore, the spectral clustering method is very sensitive to some isolated samples or noise samples, resulting in a large clustering deviation, which in turn affects the collection of face datasets.

Images