Method and system for training matting neural network, and method and device for generating portrait to be used on identity document

A computer-implemented method for creating a portrait to be used on an identity document, comprising:

• obtaining an input image comprising an area representing a face of a person;
• generating a standardized image conforming to a template by adjusting the input image, wherein a relationship between a size of the adjusted input image and a distance between target points in the adjusted input image satisfies a condition determined by the template, the target points represent parts of the person, in the adjusted input image the area is in a position defined by the template, and the template prescribes that the standardized image should depict a human face;
• inputting the standardized image to a matting neural network trained by using training images conforming to the template;
• obtaining a matte generated by the matting neural network for the standardized image;
• and
• creating the portrait based on the matte.

A computing system for training a matting neural network, comprising:

• one or more computer-readable media storing instructions;
• and
• one or more processors configured to execute the instructions to:
• obtain standardized images conforming to a template prescribing that each standardized image should depict a human face;
• and
• train the matting neural network to generate mattes for images by inputting the standardized images into the matting neural network as training images so that portraits to be used on identity documents can be created based on mattes generated by the trained matting neural network;
• wherein in obtaining the standardized images, the one or more processors are configured to:
• receive an input image comprising an area representing a face of a person;
• and
• generate one of the standardized images by adjusting the input image, wherein a relationship between a size of the adjusted input image and a distance between target points in the adjusted input image satisfies a condition determined by the template, the target points represent parts of the person, and in the adjusted input image the area is in a position defined by the template.

A computing device for creating a portrait to be used on an identity document, comprising:

• one or more computer-readable media storing instructions;
• and
• one or more processors configured to execute the instructions to:
• obtain an input image comprising an area representing a face of a person;
• generate a standardized image conforming to a template by adjusting the input image, wherein a relationship between a size of the adjusted input image and a distance between target points in the adjusted input image satisfies a condition determined by the template, the target points represent parts of the person, in the adjusted input image the area is in a position defined by the template, and the template prescribes that the standardized image should depict a human face;
• input the standardized image to a matting neural network trained by using training images conforming to the template;
• obtain a matte generated by the matting neural network for the standardized image;
• and
• create the portrait based on the matte.