Method and system to detect a text from multimedia content captured at a scene
Detection of textual phrases in a non-horizontal orientation at a scene is a critical problem. This disclosure relates to a processor implemented method to detect a text from multimedia content captured at a scene. An input original image is processed by a trained model to obtain an individual character with a bounding box on the original image. The original image is positioned by a gradient to obtain a rotated image if a number of detected characters is not equal to a number of expected characters on the original image. At least one missing character bounding box on the original image and on the rotated image are estimated to construct a horizontal text image if the number of detected characters is not equal to the number of expected characters on the rotated image. At least one missing character in the estimated bounding box is detected by at least one text returned from an optical character reader.
1. A processor implemented method, comprising:
receiving, via one or more hardware processors, an original image captured from a scene as an input;
processing, via the one or more hardware processors, the original image by a trained model to obtain at least one individual character associated with at least one bounding box on the original image;
determining, via the one or more hardware processors, whether a number of detected characters is equal to a number of expected characters (N) on the original image based on the at least one individual character associated with the at least one bounding box on the original image, wherein performing, in response to determining that the number of detected characters is not equal to the number of expected characters (N) on the original image, the steps of:
determining, via the one or more hardware processors, a gradient at which one or more texts are inclined by the at least one bounding box from the original image;
positioning, via the one or more hardware processors, the original image by the gradient to obtain a rotated image, and wherein the rotated image corresponds to the one or more texts aligned in a vertical orientation; and
processing, via the one or more hardware processors, the rotated image by the trained model to obtain at least one individual character associated with at least one bounding box on the rotated image;
determining, via the one or more hardware processors, whether a number of detected characters is equal to a number of expected characters (N) on the rotated image based on the at least one individual character associated with the at least one bounding box on the rotated image, wherein performing, in response to determining that the number of detected characters is not equal to the number of expected characters (N) on the rotated image, the steps of:
estimating, via the one or more hardware processors, one or more missing character bounding boxes on the original image, and one or more missing character bounding boxes on the rotated image; and
constructing, via the one or more hardware processors, a horizontal text image; and
detecting, via the one or more hardware processors, one or more missing characters in the one or more estimated missing character bounding boxes based on one or more returned texts.
2. The processor implemented method of claim 1 , wherein the step of detecting the one or more missing characters is terminated when the number of detected characters is equal to the number of expected characters (N) on the original image.
3. The processor implemented method of claim 1 , wherein the step of detecting the one or more missing characters is terminated when the number of detected characters is equal to the number of expected characters (N) on the rotated image.
4. The processor implemented method of claim 1 , wherein the missing character bounding boxes on the original image corresponds to: (i) a top missing detection, (ii) a middle missing detection, and (iii) a bottom missing detection, and wherein the missing character bounding boxes on the rotated image corresponds to: (i) a top missing detection, (ii) a middle missing detection, and (iii) a bottom missing detection.
5. The processor implemented method of claim 1 , wherein the horizontal text image is constructed when the at least one bounding box from the trained model on the original image and the estimated missing character bounding boxes on the original image are aligned in a horizontal orientation.
6. The processor implemented method of claim 1 , wherein the horizontal text image is constructed when the at least one bounding box from the trained model on the rotated image and the estimated missing character bounding boxes on the rotated image are aligned in a horizontal orientation.
7. A system, comprising:
a memory storing instructions;
one or more communication interfaces; and
one or more hardware processors coupled to the memory via the one or more communication interfaces,
wherein the one or more hardware processors are configured by the instructions to:
receive, an original image captured from a scene as an input;
process, the original image by a trained model to obtain at least one individual character associated with at least one bounding box on the original image;
determine, whether a number of detected characters is equal to a number of expected characters (N) on the original image based on the at least one individual character associated with the at least one bounding box on the original image, wherein perform, in response to determine that the number of detected characters is not equal to the number of expected characters (N) on the original image, the steps of:
determine, a gradient at which one or more texts are inclined by the at least one bounding box from the original image;
position, the original image by the gradient to obtain a rotated image, and wherein the rotated image corresponds to the one or more texts aligned in a vertical orientation; and
process, the rotated image by the trained model to obtain at least one individual character associated with at least one bounding box on the rotated image;
determine, whether a number of detected characters is equal to a number of expected characters (N) on the rotated image based on the at least one individual character associated with the at least one bounding box on the rotated image, wherein perform, in response to determining that the number of detected characters is not equal to the number of expected characters (N) on the rotated image, the steps of:
estimate, one or more missing character bounding boxes on the original image, and one or more missing character bounding boxes on the rotated image; and
construct, a horizontal text image and feed the horizontal image to an optical character reader (OCR); and
detect, one or more missing characters in the one or more estimated missing character bounding boxes based on one or more texts returned from the optical character reader (OCR).
8. The system of claim 7 , wherein the one or more hardware processors are further configured by the instructions to: terminate, the step of detection of the one or more missing characters when the number of detected characters is equal to the number of expected characters (N) on the original image.
9. The system of claim 7 , wherein the one or more hardware processors are further configured by the instructions to: terminate, the step of detection of the one or more missing characters when the number of detected characters is equal to the number of expected characters (N) on the rotated image.
10. The system of claim 7 , wherein the missing character bounding boxes on the original image corresponds to: (i) a top missing detection, (ii) a middle missing detection, and (iii) a bottom missing detection, and wherein the missing character bounding boxes on the rotated image corresponds to: (i) a top missing detection, (ii) a middle missing detection, and (iii) a bottom missing detection.
11. The system of claim 7 , wherein the horizontal text image is constructed when the at least one bounding box from the trained model on the original image and the estimated missing character bounding boxes on the original image are aligned in a horizontal orientation.
12. The system of claim 7 , wherein the horizontal text image is constructed when the at least one bounding box from the trained model on the rotated image and the estimated missing character bounding boxes on the rotated image are aligned in a horizontal orientation.
13. One or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause:
receiving, an original image captured from a scene as an input;
processing, the original image by a trained model to obtain at least one individual character associated with at least one bounding box on the original image;
determining, whether a number of detected characters is equal to a number of expected characters (N) on the original image based on the at least one individual character associated with the at least one bounding box on the original image, wherein performing, in response to determining that the number of detected characters is not equal to the number of expected characters (N) on the original image, the steps of:
determining, a gradient at which one or more texts are inclined by the at least one bounding box from the original image;
positioning, the original image by the gradient to obtain a rotated image, and wherein the rotated image corresponds to the one or more texts aligned in a vertical orientation; and
processing, the rotated image by the trained model to obtain at least one individual character associated with at least one bounding box on the rotated image;
determining, whether a number of detected characters is equal to a number of expected characters (N) on the rotated image based on the at least one individual character associated with the at least one bounding box on the rotated image, wherein performing, in response to determining that the number of detected characters is not equal to the number of expected characters (N) on the rotated image, the steps of:
estimating, one or more missing character bounding boxes on the original image, and one or more missing character bounding boxes on the rotated image; and
constructing, a horizontal text image; and
detecting, one or more missing characters in the one or more estimated missing character bounding boxes based on one or more returned texts.
14. The one or more non-transitory machine-readable information storage mediums of claim 13 , wherein the step of detecting the one or more missing characters is terminated when the number of detected characters is equal to the number of expected characters (N) on the original image.
15. The one or more non-transitory machine-readable information storage mediums of claim 13 , wherein the step of detecting the one or more missing characters is terminated when the number of detected characters is equal to the number of expected characters (N) on the rotated image.
16. The one or more non-transitory machine-readable information storage mediums of claim 13 , wherein the missing character bounding boxes on the original image corresponds to: (i) a top missing detection, (ii) a middle missing detection, and (iii) a bottom missing detection, and wherein the missing character bounding boxes on the rotated image corresponds to: (i) a top missing detection, (ii) a middle missing detection, and (iii) a bottom missing detection.
17. The one or more non-transitory machine-readable information storage mediums of claim 13 , wherein the horizontal text image is constructed when the at least one bounding box from the trained model on the original image and the estimated missing character bounding boxes on the original image are aligned in a horizontal orientation.
18. The one or more non-transitory machine-readable information storage mediums of claim 13 , wherein the horizontal text image is constructed when the at least one bounding box from the trained model on the rotated image and the estimated missing character bounding boxes on the rotated image are aligned in a horizontal orientation.