Granted Patent B2

US 12,374,074 · App. 17/962,170 · Granted Jul 29, 2025

Inventors: Abhishek Mishra (Noida, IN); Sai Hemanth Kasaraneni (Noida, IN); Saurabh Kumar (Noida, IN)

Assignee: Samsung Electronics Co., Ltd.

G06V10/25G06T3/40G06T7/50G06T7/60G06T7/70G06V10/764G06V10/86H04N5/2628G06T2207/20076G06T2207/20081G06T2207/20084

Abstract

Claims (90)

1. A method for automatically generating a Region Of Interest (ROI) centric image performed by an electronic device, the method comprising:

receiving, by the electronic device, at least one image frame, wherein the at least one image frame comprises a plurality of objects;

identifying, by the electronic device, at least one first ROI, at least one second ROI, and at least one non-ROI in the at least one image frame, wherein the identifying includes:

identifying the at least one first ROI based on an object importance score,

identifying the at least one second ROI based on an object of interest of the at least one first ROI, and

identifying the at least one non-ROI based on a highest redundancy score and the object importance score;

rescaling, by the electronic device, each ROI in the at least one image frame;

summarizing, by the electronic device, the at least one non-ROI in the at least one image frame, wherein the summarizing includes compressing repetitive object fragments in the at least one non-ROI; and

automatically generating, by the electronic device, the ROI centric image based on the rescaling and the summarizing.

2. The method of claim 1 , further comprising:

detecting, by the electronic device, the plurality of objects presented in the at least one image frame, wherein the plurality of objects comprises three-dimensional (3D) location information;

determining, by the electronic device, two-dimensional (2D) coordinates information of the plurality of detected objects, wherein the 2D coordinates information comprises center information, height information, width information, and tag information, wherein the 2D coordinates information is detected by at least one Machine Learning (ML) model;

identifying, by the electronic device, a depth of the plurality of detected objects using at least one camera of the electronic device;

determining, by the electronic device, a relative distance between each object of the plurality of detected objects using a triangulation mechanism; and

identifying, by the electronic device, the 3D location information of the plurality of objects presented in the at least one image frame along with object categorization by using the 2D coordinates information, the identified depth of the plurality of detected objects, and the determined relative distance between each object,

wherein the tag information is used to categorize each object.

3. The method of claim 1 , wherein the identifying of the at least one second ROI based on the object of interest of the at least one first ROI comprises:

determining, by the electronic device, a probability of interest of the at least one first ROI on each object based on a plurality of characteristics of each object;

ranking, by the electronic device, the determined probability; and

selecting, by the electronic device, at least one object having a highest rank, wherein the at least one object is used as the at least one second ROI to rescale.

4. The method of claim 3 ,

wherein the plurality of characteristics comprises:

relative position information between each object,

at least one surface of at least one object oriented towards at least one surface of at least one other object,

an extension of or a front surface of the at least one object oriented towards the at least one other object,

a relationship between each object, and

user-specific parameters, and

wherein the user-specific parameters comprise an image capturing pattern or history of a user of the electronic device.

5. The method of claim 1 , further comprising:

generating, by the electronic device, a scene graph associated with the plurality of objects;

applying, by the electronic device, a relation classifier on the generated scene graph; and

determining, by the electronic device, a relationship between each pair of objects of the plurality of objects using the relation classifier.

6. The method of claim 1 , wherein the rescaling, by the electronic device, of each ROI in the at least one image frame comprises:

extracting, by the electronic device, a relationship between each pair of objects of the plurality of objects;

determining, by the electronic device, an actual size of each object presented in the at least one image frame and an actual Z-order of each object presented in the at least one image frame;

generating, by the electronic device, a scene graph based on the plurality of detected objects and the actual size of each object;

generating, by the electronic device, a textual knowledge graph based on the extracted relationship between each object and the actual Z-order of each object; and

rescaling, by the electronic device, the actual size of each ROI and the actual Z-order of each ROI by using the generated scene graph and the generated textual knowledge graph.

7. The method of claim 1 , wherein the summarizing, by the electronic device, of the at least one non-ROI in the at least one image frame comprises:

determining, by the electronic device, an object importance score associated with each object;

masking, by the electronic device, each object based on 3D location information;

determining, by the electronic device, the at least one non-ROI based on the object importance score;

determining, by the electronic device, a highest redundancy score associated with the at least one non-ROI, wherein the highest redundancy score is determined by at least one Machine Learning (ML) model; and

summarizing, by the electronic device, the at least one non-ROI in the at least one image frame, wherein the at least one non-ROI has the highest redundancy score.

8. The method of claim 1 , wherein the automatically generating, by the electronic device, the ROI centric image comprises:

identifying, by the electronic device, the plurality of objects in the at least one image frame, wherein the plurality of objects comprises the at least one rescaled-first ROI, the at least one rescaled-second ROI, and the at least one summarized non-ROI;

stitching, by the electronic device, the identified plurality of objects to generate the ROI centric image; and

automatically generating, by the electronic device, the ROI centric image.

9. An electronic device for automatically generating a Region Of Interest (ROI) centric image, the electronic device comprising:

memory storing one or more computer programs;

one or more processors; and

a three-dimensional (3D) planar image controller, operably connected to the memory and the processor,

wherein the one or more computer programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the electronic device to:

receive at least one image frame, wherein the at least one image frame comprises a plurality of objects,

identify at least one first ROI, at least one second ROI, and at least one non-ROI in the at least one image frame,

rescale each ROI in the at least one image frame,

summarize the at least one non-ROI in the at least one image frame,

wherein, to summarize the at least one non-ROI, the one or more computer programs further include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the electronic device to compress repetitive object fragments in the at least one non-ROI, and

automatically generate the ROI centric image based on each ROI being rescaled and the at least one non-ROI being summarized,

wherein the electronic device further comprising:

an Object Of Interest (OOI) detector configured to identify the at least one first ROI based on an object importance score, the at least one second ROI based on an object of interest of the at least one first ROI, and the at least one non-ROI based on a highest redundancy score and the object importance score.

10. The electronic device of claim 9 further comprising:

a scene identifier of the 3D planar image controller configured to:

detect the plurality of objects presented in the at least one image frame, wherein the plurality of objects comprises three-dimensional (3D) location information,

determine two-dimensional (2D) coordinates information of the plurality of detected objects, wherein the 2D coordinates information comprises center information, height information, width information, and tag information, and wherein the 2D coordinates information is detected by at least one Machine Learning (ML) model,

identify a depth of the plurality of detected objects using at least one camera of the electronic device,

determine a relative distance between each object of the plurality of detected objects using a triangulation mechanism, and

identify the 3D-location information of the plurality of objects presented in the at least one image frame along with object categorization by using the 2D coordinates information, the identified depth of the plurality of detected objects, and the determined relative distance between each object,

wherein the tag information is used to categorize each object.

11. The electronic device of claim 9 , wherein the ML model comprises a CenterNet model.

12. The electronic device of claim 9 , wherein the ML model comprises a You only look once (YOLO) model.

13. The electronic device of claim 9 further comprising:

an Object Of Interest (OOI) detector of the 3D planar image controller configured to:

determine a probability of interest of the at least one first ROI on each object based on a plurality of characteristics of each object,

rank the determined probability, and

select at least one object having a highest rank, wherein the at least one object used as the at least one second ROI to rescale.

14. The electronic device of claim 13 ,

wherein the plurality of characteristics comprises:

relative position information between each object,

at least one surface of at least one object oriented towards at least one surface of at least one other object,

an extension of or a front surface of the at least one object oriented towards the at least one other object,

a relationship between each object, and

user-specific parameters, and

wherein the user-specific parameters comprise an image capturing pattern or history of a user of the electronic device.

15. The electronic device of claim 9 , further comprising:

an object relation detector,

wherein the object relation detector is configured to:

generate a scene graph associated with the plurality of objects,

apply a relation classifier on the generated scene graph, and

determine a relationship between each pair of objects of the plurality of objects using the relation classifier.

Assignments (1)

Priority Claims (1)

Continuity (2)

References Cited (37)