Method for processing two-dimensional image and device for executing method
AssigneeSamsung Electronics Co., Ltd.
InventorsIvan Viktorovich Glazistov, Ivan Olegovich Karacharov, Andrey Yurievich Shcherbinin, Ilya Vasilievich Kurilin
Life of this patent
Estimating the term…
Abstract
Provided is a two-dimensional (2D) image processing method including obtaining a 2D image, processing the obtained 2D image by using a trained convolutional neural network (CNN) to obtain at least one camera parameter and at least one face model parameter from the 2D image, and generating a three-dimensional (3D) face model, based on the obtained at least one camera parameter and at least one face model parameter.
Claims13
A two-dimensional (2D) image processing method comprising:
- obtaining a 2D image;
- processing the obtained 2D image by using a trained convolutional neural network (CNN) to obtain at least one camera parameter, at least one face model parameter and at least one emotion parameter from the 2D image;
- generating a three-dimensional (3D) face model, based on the obtained at least one camera parameter and at least one face model parameter;
- determining a feeling of a user by using the obtained at least one emotion parameter;
- and
- providing an advertisement of a product or service, based on the determined feeling of the user,
- wherein the obtained at least one emotion parameter is based on predefined landmarks of the 2D image compared to a 3D morphable face model, and
- wherein the processing of the 2D image by using the trained CNN comprises:
- adjusting a size of the 2D image,
- detecting a face from the size-adjusted 2D image,
- indicating, with a bounding box, a face region comprising the detected face,
- cutting an image of the face region indicated by the bounding box, from the 2D image,
- generating a second image by adjusting a size of the cut image of the face region,
- marking at least one landmark of the face on the second image, and
- obtaining at least one camera parameter and at least one face model parameter that most closely match the at least one landmark.
A two-dimensional (2D) image processing device comprising:
- a memory storing one or more instructions;
- and
- a processor configured to execute the one or more instructions,
- wherein the processor is configured to:
- obtain a 2D image,
- process the obtained 2D image by using a trained convolutional neural network (CNN) to obtain at least one camera parameter, at least one face model parameter and at least one emotion parameter from the 2D image,
- generate a three-dimensional (3D) face model, based on the obtained at least one camera parameter and at least one face model parameter,
- determine a feeling of a user by using the obtained at least one emotion parameter, and
- provide an advertisement of a product or service, based on the determined feeling of the user,
- wherein the obtained at least one emotion parameter is based on predefined landmarks of the 2D image compared to a 3D morphable face model, and
- wherein the processor is further configured to:
- process the 2D image by using the trained CNN by adjusting a size of the 2D image,
- detect a face from the size-adjusted 2D image,
- indicate, with a bounding box, a face region comprising the detected face
- cut an image of the face region indicated by the bounding box, from the 2D image,
- generate a second image by adjusting a size of the cut image of the face region,
- mark at least one landmark of the face on the second image, and
- obtain at least one camera parameter and at least one face model parameter that most closely match the at least one landmark.
Dependencies are inferred from the claim text.