With the rapid development of artificial intelligence technology, data has become the main factor in various artificial intelligence applications. From behavior monitoring to image recognition, the performance of artificial intelligence systems is highly dependent on the quality and diversity of data sets. However, in the face of massive data demands, how to collect and manage this data remains a huge challenge.

The face recognition system is composed of two parts: hardware and software. The hardware is mainly the camera and the computer. The software of the face recognition system controls the camera to collect pictures, preprocess the collected pictures, and then perform the tasks of face detection and positioning, face feature extraction and face feature matching.

When the mask covers most areas of the face, the facial features available for face recognition are greatly reduced. The face recognition system cannot extract the complete facial features of the face, but part of the face information. If the face picture saved in the system is a complete face without a mask, the difference between the two pictures will be large, the face recognition system will fail to verify the identity.

In order to ensure that the masked face can be correctly recognized, the face recognition system should be able to detect and locate the masked face. Generally speaking, this problem can be solved by pre-training a masked face recognition model. For example, the face recognition system uses Gaussian masking method to change the weight distribution mode of network feature training when training the face recognition model, and increases the proportion of the non-occlusive part of the face and the head part, so that the model can detect and locate the masked face.

After detecting and locating the face, different templates used for facial feature comparison are determined. If the face is not wearing a mask, the standard template library is used to compare the recognition results according to the conventional process; if the face is wearing a mask, the feature attention mask is used in the feature extraction process to obtain the people outside the part of the mask. The facial information features are then compared with the mask-wearing template library, and the identity verification result is output according to the comparison result.

Masked Face Recognition Data Solution

Nexdata has developed the “4,608 People — Face Recognition Data with Gauze Mask” and “1,031 People with Occlusion and Multi-pose Face Recognition Data”, to support the innovation of face recognition. Nexdata strictly abides by the relevant regulations, and the data is collected with proper data collection authorization agreement. Nexdata is committed to promoting the innovation of emotion recognition technology with high-quality data and fully ensuring the data security.

4,608 People — Face Recognition Data with Gauze Mask

4,608 people, 7 images per person. Age ranges from teenager to the elderly. The accuracy of labels of mask type, gender, race and age is over 97%.

1,031 People with Occlusion and Multi-pose Face Recognition Data

1,031 people, 36 images per person. Age ranges from teenager to the elderly. The accuracy of labels of occlusion type, gender, race and age is over 97%.

End

If you need data services, please feel free to contact us: [email protected]

With the continuous advance of data technology, we can look expect more innovative AI applications emerge in all walks of life. As we mentioned at the beginning, the importance of data in AI cannot be ignored, and high-quality data will continuously drive technological breakthroughs.