Device for automatically capturing photo or video about specific moment, and operation method thereof

1. A method of automatically obtaining a photo of a moment, the method comprising:

obtaining a plurality of image frames sequentially captured through a camera for a preset time interval;

detecting at least one image frame among the plurality of image frames in which a main object corresponding to a preset main object or an action of the main object corresponding to a preset action is recognized;

determining a type of composition of the at least one image frame;

cropping a region including the main object from the at least one image frame based on placement of objects in the determined type of composition;

predicting an aesthetic estimation score of a cropped image frame including the cropped region by inputting the cropped image frame into a first deep neural network model;

adjusting at least one of image quality parameters of the cropped image frame based on the predicted aesthetic estimation score; and

obtaining the photo using the cropped image frame of which the at least one of the image quality parameters is adjusted.

2. The method of claim 1 , wherein the obtaining the photo comprises:

adjusting the at least one of the image quality parameters including brightness, definition, chroma, contrast, or high dynamic range (HDR) of the cropped image frame.

3. The method of claim 1 , wherein the determining the type of composition of the at least one image frame comprises determining the type of composition of the at least one image frame by using a second deep neural network model obtained by training the second deep neural network model with an input of a plurality of photos to output label values of types of composition.

4. The method of claim 1 , wherein the first deep neural network model is trained with a plurality of photos as an input and normalized values of aesthetic scores estimated in advance for the plurality of photos as an output.

5. The method of claim 2 , wherein the adjusting at least one of the image quality parameters of the cropped image frame by using a third deep neural network model obtained by training a model parameter which indicates a loss between an original photo and the image frame.

6. The method of claim 1 , further comprising displaying the obtained photo,

wherein the display of the photo comprises displaying at least one user interface of a guideline of a composition, a histogram of types of composition, or an aesthetic prediction score by overlaying the at least one user interface on the photo.

7. A device for automatically obtaining a photo of a moment, the device comprising:

a camera configured to sequentially capture a plurality of image frames of a subject for a preset time interval;

a storage storing the plurality of image frames;

a memory storing a program including one or more instructions; and

a processor configured to execute the one or more instructions of the program stored in the memory to:

detect at least one image frame among the plurality of image frames in which a main object corresponding to a preset main object or an action of the main object corresponding to a preset action is recognized;

determine a type of composition of the at least one image frame;

crop a region including the main object from the at least one image frame based on placement of objects in the determined type of composition;

predict an aesthetic estimation score of a cropped image frame including the cropped region by inputting the cropped image frame into a first deep neural network model;

adjust at least one of image quality parameters of the cropped image frame based on the predicted aesthetic estimation score;

obtain the photo using the cropped image frame of which the at least one of the image quality parameters is adjusted; and

store the obtained photo in the storage.

8. The device of claim 7 , wherein the processor is further configured to adjust the at least one of the image quality parameters including brightness, definition, chroma, contrast, or high dynamic range (HDR) of the cropped image frame.

9. The device of claim 7 , wherein the processor is further configured to determine the type of composition of the at least one image frame by using a second deep neural network model obtained by training through the second deep neural network model with an input of a plurality of photos to output label values of types of composition.

10. The device of claim 9 , wherein the first deep neural network model is trained with a plurality of photos as an input and normalized values of aesthetic scores estimated in advance for the plurality of photos as an output.

11. The device of claim 8 , wherein the processor is further configured to adjust the at least one of the image quality parameters of the cropped image frame by using a third deep neural network model obtained by training a model parameter which indicates a loss between an original photo and the adjusted image frame.

12. The device of claim 7 , further comprising a display configured to display the obtained photo,

wherein the processor is further configured to control the display to display at least one user interface of a guideline of a composition, a histogram of types of composition, or an aesthetic prediction score by overlaying the at least one user interface on the photo.

13. A non-transitory computer-readable storage medium storing a computer program, the computer program, when executed by a processor, causes the processor to:

obtain a plurality of image frames sequentially captured through a camera for a preset time interval;

detect at least one image frame among the plurality of image frames in which a main object corresponding to a preset main object or an action of the main object corresponding to a preset action is recognized;

determine a type of composition of the at least one image frame;

crop a region including the main object from the at least one image frame based on placement of objects in the at least one image frame and the determined type of composition;

predict an aesthetic estimation score of a cropped image frame including the cropped region by inputting the cropped image frame into a deep neural network model;

adjust at least one of image quality parameters of the cropped image frame based on the predicted aesthetic estimation score; and

obtain a photo using the cropped image frame of which the at least one of the image quality parameters is adjusted.