Rapid language detection for characters in images of documents

1. A computer-implemented method for rapid language detection of documents, comprising:

receiving an image of a document having characters that correspond to a given language;

using a text recognition algorithm to determine a first language believed to correspond to the characters in the image of the document by:

using a generated mixed language corpus to convert an input sequence of the characters in the image of the document into one or more n-gram sequences, and

computing a hidden layer vector for the one or more n-gram sequences,

wherein the text recognition algorithm is trained using a simple shallow neural network and the generated mixed language corpus,

wherein the generated mixed language corpus is formed by:

randomly sampling one or more libraries having vocabulary and/or characters therein, and

combining the randomly sampled vocabulary and/or characters from the one or more libraries to form the generated mixed language corpus;

computing a first confidence level associated with the first language believed to correspond to the characters in the image of the document;

determining whether the first confidence level associated with the first language is outside a predetermined range; and

in response to determining that the first confidence level associated with the first language is not outside the predetermined range, outputting the first language as the given language.

2. The computer-implemented method of claim 1 , wherein using a text recognition algorithm to determine a first language believed to correspond to the characters in the image of the document includes:

mapping the hidden layer vector to a language label.

3. The computer-implemented method of claim 2 , wherein computing a hidden layer vector for the one or more n-gram sequences includes:

calculating an average of the n-gram sequences.

4. The computer-implemented method of claim 1 , comprising:

in response to determining that the first confidence level associated with the first language is outside the predetermined range, computing a second confidence level associated with a second language believed to correspond to the characters in the image of the document;

determining whether the second confidence level associated with the second language is outside the predetermined range; and

in response to determining that the second confidence level associated with the second language is not outside the predetermined range, outputting the second language believed to correspond to the characters in the image of the document as the given language.

5. The computer-implemented method of claim 4 , comprising:

in response to determining that the second confidence level associated with the second language is outside the predetermined range, computing a third confidence level associated with a third language believed to correspond to the characters in the image of the document;

determining whether the third confidence level associated with the third language is outside the predetermined range; and

in response to determining that the third confidence level associated with the third language is not outside the predetermined range, outputting the third language believed to correspond to the characters in the image of the document as the given language.

6. The computer-implemented method of claim 1 , with a proviso that the generated mixed language corpus does not have semantic coherence.

7. The computer-implemented method of claim 1 , wherein the simple shallow neural network is a long short-term memory (LSTM).

8. The computer-implemented method of claim 1 , wherein the simple shallow neural network is a support vector machine.

9. The computer-implemented method of claim 1 , comprising:

in response to outputting the first language as the given language, sending one or more instructions to apply an optical character recognition (OCR) model to the image of the document, wherein the OCR model corresponds to the first language.

10. A computer program product for rapid language detection of documents, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable and/or executable by a processor to cause the processor to:

receive, by the processor, an image of a document having characters that correspond to a given language;

use, by the processor, a text recognition algorithm to determine a first language believed to correspond to the characters in the image of the document by:

using a generated mixed language corpus to convert an input sequence of the characters in the image of the document into one or more n-gram sequences, and

computing a hidden layer vector for the one or more n-gram sequences,

wherein the text recognition algorithm is trained using a simple shallow neural network and the generated mixed language corpus,

wherein the generated mixed language corpus is formed by:

randomly sampling one or more libraries having vocabulary and/or characters therein, and

combining the randomly sampled vocabulary and/or characters from the one or more libraries to form the generated mixed language corpus;

compute, by the processor, a first confidence level associated with the first language believed to correspond to the characters in the image of the document;

determine, by the processor, whether the first confidence level associated with the first language is outside a predetermined range; and

in response to determining that the first confidence level associated with the first language is not outside the predetermined range, output, by the processor, the first language believed to correspond to the characters in the image of the document as the given language.

11. The computer program product of claim 10 , wherein using a text recognition algorithm to determine a first language believed to correspond to the characters in the image of the document includes:

mapping the hidden layer vector to a language label.

12. The computer program product of claim 11 , wherein computing a hidden layer vector for the one or more n-gram sequences includes: calculating an average of the n-gram sequences.

13. The computer program product of claim 10 , wherein the program instructions are readable and/or executable by the processor to cause the processor to:

in response to determining that the first confidence level associated with the first language is outside the predetermined range, compute, by the processor, a second confidence level associated with a second language believed to correspond to the characters in the image of the document;

determine, by the processor, whether the second confidence level associated with the second language is outside the predetermined range; and

in response to determining that the second confidence level associated with the second language is not outside the predetermined range, output, by the processor, the second language believed to correspond to the characters in the image of the document as the given language.

14. The computer program product of claim 13 , wherein the program instructions are readable and/or executable by the processor to cause the processor to:

in response to determining that the second confidence level associated with the second language is outside the predetermined range, compute, by the processor, a third confidence level associated with a third language believed to correspond to the characters in the image of the document;

determine, by the processor, whether the third confidence level associated with the third language is outside the predetermined range; and

in response to determining that the third confidence level associated with the third language is not outside the predetermined range, output, by the processor, the third language believed to correspond to the characters in the image of the document as the given language.

15. The computer program product of claim 10 , with a proviso that the generated mixed language corpus does not have semantic coherence.

16. The computer program product of claim 10 , wherein the simple shallow neural network is a long short-term memory (LSTM).

17. The computer program product of claim 10 , wherein the simple shallow neural network is a support vector machine.

18. A system, comprising:

a processor; and

logic integrated with the processor, executable by the processor, or integrated with and executable by the processor, the logic being configured to:

receive, by the processor, an image of a document having characters that correspond to a given language;

use, by the processor, a text recognition algorithm to determine a first language believed to correspond to the characters in the image of the document by:

using a generated mixed language corpus to convert an input sequence of the characters in the image of the document into one or more n-gram sequences, and

computing a hidden layer vector for the one or more n-gram sequences, wherein the text recognition algorithm is trained using a simple shallow neural network and the generated mixed language corpus,

wherein the generated mixed language corpus is formed by:

randomly sampling one or more libraries having vocabulary and/or characters therein, and

combining the randomly sampled vocabulary and/or characters from the one or more libraries to form the generated mixed language corpus;

compute, by the processor, a first confidence level associated with the first language believed to correspond to the characters in the image of the document;

determine, by the processor, whether the first confidence level associated with the first language is outside a predetermined range; and

in response to determining that the first confidence level associated with the first language is not outside the predetermined range, output, by the processor, the first language believed to correspond to the characters in the image of the document as the given language.

19. The system of claim 18 , wherein using a text recognition algorithm to determine a first language believed to correspond to the characters in the image of the document includes:

mapping the hidden layer vector to a language label,

wherein computing a hidden layer vector for the one or more n-gram sequences includes: calculating an average of the n-gram sequences.

20. The system of claim 18 , with a proviso that the generated mixed language corpus does not have semantic coherence, wherein the simple shallow neural network is a long short-term memory (LSTM).