Automated document classification
First data corresponding to a first document is obtained. A first feature vector is generated for the first data. The first feature vector is provided as an input to a classifier. The classifier is trained to characterize the first document as having one of a first characteristic or a second characteristic based on the first feature vector.
1. A method comprising:
obtaining, by a data acquisition module of a hardware computing system, first data corresponding to a first document, the first data comprising pixel-level patterns for a document image that represents the first document, the pixel-level patterns comprising a cell having multiple image pixels;
computing, at the hardware computing system, mathematical relationships between each of the multiple image pixels using a threshold constant to threshold each image pixel into multiple values to generate a range of values;
generating, using a processing device of the hardware computing system, a first feature vector for the first data based on binary data that represents the range of values generated from the computed mathematical relationships between each of the multiple image pixels;
providing, using the processing device, the first feature vector as an input to a classifier of the hardware computing system, the input being used for training the classifier; and
training the classifier to characterize the first document as having one of a first characteristic or a second characteristic based on the first feature vector.
2. The method of claim 1 , wherein the first characteristic corresponds to identifying a document as real, and the second characteristic corresponds to identifying a document as fake.
3. The method of claim 1 , further comprising:
iteratively providing, as inputs to the classifier, feature vectors associated with a plurality of documents for training the classifier, wherein each document in the plurality has one of the first characteristic or the second characteristic.
4. The method of claim 1 , further comprising:
obtaining second data corresponding to a second document;
generating a second feature vector for the second data;
providing the second feature vector as an input to the classifier;
estimating, by the classifier and based on the second feature vector, whether the second document corresponds to the first characteristic or the second characteristic; and
outputting, by the classifier, a measure of the estimate whether the second document corresponds to the first characteristic or the second characteristic.
5. The method of claim 4 , wherein estimating whether the second document corresponds to the first characteristic or the second characteristic comprises:
categorizing, by the classifier and based on the training, a plurality of feature vectors into one of a first group corresponding to the first characteristic or a second group corresponding to the second characteristic;
determining, by the classifier, whether the second feature vector corresponds to feature vectors included in the first group or the second group;
estimating, by the classifier, that the second document corresponds to the first characteristic conditioned on determining that the second feature vector corresponds to feature vectors included in the first group; and
estimating, by the classifier, that the second document corresponds to the second characteristic conditioned on determining that the second feature vector corresponds to feature vectors included in the second group.
6. The method of claim 4 , wherein obtaining the first data includes obtaining an image of the first document and obtaining the second data includes obtaining an image of the second document.
7. The method of claim 6 , wherein generating the first feature vector comprises generating a pixel-level pattern of the image the first document, the method further comprising:
dividing the first data into cells, each cell including multiple image pixels;
for each pixel in the multiple image pixels in a cell:
comparing the pixel to a group of neighboring pixels,
for each neighboring pixel in the group, recording a first value based on determining that a value of the pixel is greater than a value of the neighboring pixel, and recording a second value based on determining that the value of the pixel is smaller than the value of the neighboring pixel, and
computing a number for the pixel that is composed of one or more of first values and second values determined for the group of neighboring pixels;
for each cell, generating a histogram of the numbers for the pixels included in the cell;
aggregating the histograms for all the cells corresponding to the first data; and
generating the first feature vector for the first data based on the aggregated histograms.
8. The method of claim 7 , wherein the number computed for a pixel from the first value and the second value corresponds to a binary value.
9. The method of claim 7 , wherein the number for each pixel includes a binary number with a number of digits that corresponds to a computed number from the neighboring pixels in the group of neighboring pixels.
10. The method of claim 7 , wherein generating the histogram for each cell further comprises normalizing the histogram.
11. The method of claim 6 , wherein generating the first feature vector or the second feature vector comprises generating one or more of local ternary patterns, image texture patterns, or image pattern combinations and variations, wherein the first feature vector or the second feature vector is generated based on the one or more local ternary patterns by computing the mathematical relationships between image pixels in a cell using the threshold constant to threshold each image pixel into multiple values.
12. The method of claim 1 , wherein the classifier includes one of a neural network classifier or a support vector machine (SVM) classifier.
13. The method of claim 1 , wherein the first feature vector includes a collection of numbers derived from the document or the binary data that represents the range of values generated from the computed mathematical relationships between each of the multiple image pixels.
14. The method of claim 1 , wherein the document includes a physical object with information printed, written, etched or pressed by one or more fabricating means.
15. The method of claim 1 , wherein the document is a digital document with information and layout rendered on a display device, and
wherein the display device is one of a display monitor, a television, a phone, or a device configured to render and display digital information.
16. A non-transitory storage medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
obtaining, by a data acquisition module of a hardware computing system, first data corresponding to a first document, the first data comprising pixel-level patterns for a document image that represents the first document, the pixel-level patterns comprising a cell having multiple image pixels;
computing, at the hardware computing system, mathematical relationships between each of the multiple image pixels using a threshold constant to threshold each image pixel into multiple values to generate a range of values;
generating, using a processing device of the hardware computing system, a first feature vector for the first data based on binary data that represents the range of values generated from the computed mathematical relationships between each of the multiple image pixels;
providing, using the processing device, the first feature vector as an input to a classifier of the hardware computing system, the input being used for training the classifier; and
training the classifier to characterize the first document as having one of a first characteristic or a second characteristic based on the first feature vector.
17. The non-transitory storage medium of claim 16 , wherein the first characteristic corresponds to identifying a document as real, and the second characteristic corresponds to identifying a document as fake.
18. The non-transitory storage medium of claim 16 , wherein the operations further comprise:
iteratively providing, as inputs to the classifier, feature vectors associated with a plurality of documents for training the classifier, wherein each document in the plurality has one of the first characteristic or the second characteristic.
19. The non-transitory storage medium of claim 16 , wherein the operations further comprise:
obtaining second data corresponding to a second document;
generating a second feature vector for the second data;
providing the second feature vector as an input to the classifier;
estimating, by the classifier and based on the second feature vector, whether the second document corresponds to the first characteristic or the second characteristic; and
outputting, by the classifier, a measure of the estimate whether the second document corresponds to the first characteristic or the second characteristic.
20. The non-transitory storage medium of claim 19 , wherein estimating whether the second document corresponds to the first characteristic or the second characteristic comprises:
categorizing, by the classifier and based on the training, a plurality of feature vectors into one of a first group corresponding to the first characteristic or a second group corresponding to the second characteristic;
determining, by the classifier, whether the second feature vector corresponds to feature vectors included in the first group or the second group;
estimating, by the classifier, that the second document corresponds to the first characteristic conditioned on determining that the second feature vector corresponds to feature vectors included in the first group; and
estimating, by the classifier, that the second document corresponds to the second characteristic conditioned on determining that the second feature vector corresponds to feature vectors included in the second group.
21. A system comprising:
one or more processors; and
a non-transitory storage device storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
obtaining, by a data acquisition module of a hardware computing system, first data corresponding to a first document, the first data comprising pixel-level patterns for a document image that represents the first document, the pixel-level patterns comprising a cell having multiple image pixels;
computing, at the hardware computing system, mathematical relationships between each of the multiple image pixels using a threshold constant to threshold each image pixel into multiple values to generate a range of values;
generating, using a processing device of the hardware computing system, a first feature vector for the first data based on binary data that represents the range of values generated from the computed mathematical relationships between each of the multiple image pixels;
providing, using the processing device, the first feature vector as an input to a classifier of the hardware computing system, the input being used for training the classifier; and
training the classifier to characterize the first document as having one of a first characteristic or a second characteristic based on the first feature vector.
22. The system of claim 21 , wherein the first characteristic corresponds to identifying a document as real, and the second characteristic corresponds to identifying a document as fake.
23. The system of claim 21 , wherein the operations further comprise:
iteratively providing, as inputs to the classifier, feature vectors associated with a plurality of documents for training the classifier, wherein each document in the plurality has one of the first characteristic or the second characteristic.
24. The system of claim 21 , wherein the operations further comprise:
obtaining second data corresponding to a second document;
generating a second feature vector for the second data;
providing the second feature vector as an input to the classifier;
estimating, by the classifier and based on the second feature vector, whether the second document corresponds to the first characteristic or the second characteristic; and
outputting, by the classifier, a measure of the estimate whether the second document corresponds to the first characteristic or the second characteristic.
25. The system of claim 24 , wherein estimating whether the second document corresponds to the first characteristic or the second characteristic comprises:
categorizing, by the classifier and based on the training, a plurality of feature vectors into one of a first group corresponding to the first characteristic or a second group corresponding to the second characteristic;
determining, by the classifier, whether the second feature vector corresponds to feature vectors included in the first group or the second group;
estimating, by the classifier, that the second document corresponds to the first characteristic conditioned on determining that the second feature vector corresponds to feature vectors included in the first group; and
estimating, by the classifier, that the second document corresponds to the second characteristic conditioned on determining that the second feature vector corresponds to feature vectors included in the second group.