Form type learning system and image processing apparatus
To accurately classify a form without using form layout information, the image processing apparatus utilizes a classifier that accepts a filled-in form whose image has been reduced into a specific size as an input and specifies the form type of the filled-in form. Machine learning has been performed to the classifier by a form type learning system and the form type learning system reduces an image of a filled-in form as an original document image, adds a noise to the original document image, which has not been reduced or has been reduced, to generate multiple images for machine learning, associates the form type of the original document image with the multiple images for machine learning as a label, and performs machine learning of the classifier using the multiple images for machine learning and the label as training data.
1. A form type learning system comprising:
an original document image acquiring unit that acquires an image of a filled-in form as an original document image;
an image reducing unit that reduces the original document image;
a noise adding unit that adds a noise to the original document image to generate a plurality of images for machine learning, the original document image having not been reduced by the image reducing unit or having been reduced by the image reducing unit;
a label adding unit that associates a form type of the original document image with the plurality of images for machine learning as a label; and
a machine learning processing unit that performs machine learning of a classifier using the plurality of images for machine learning and the label as training data,
wherein the classifier accepts an image of a filled-in form as an input and outputs a form type.
2. The form type learning system according to claim 1 ,
wherein the noise adding unit adds a pseudo character image to the original document image as the noise to generate the plurality of images for machine learning, and
wherein the pseudo character image is in a specific shape and a specific size.
3. The form type learning system according to claim 2 , wherein the specific size is any size in a range from a minimum size to a maximum size of a character that can exist in the original document image, which has not been reduced the image reducing unit or has been reduced by the image reducing unit.
4. The form type learning system according to claim 2 , wherein the noise adding unit adds the pseudo character image of different aspect ratios to a plurality of adding positions, respectively.
5. The form type learning system according to claim 2 , wherein the noise adding unit detects a frame in the original document image and adds the pseudo character image to an inside of the detected frame.
6. The form type learning system according to claim 1 , wherein the noise adding unit randomly specifies an adding position of the noise and, in a case where a density of the specified adding position is different from that of a background, the noise adding unit changes the adding position to another position having the density of the background.
7. An image processing apparatus comprising:
a form image acquiring unit that acquires an image of a filled-in form;
an image reducing unit that reduces the acquired image of the filled-in form into a specific size; and
a classifier that accepts the reduced image of the filled-in form as an input and outputs a form type,
wherein machine learning has been performed to the classifier by a form type learning system, and
wherein the form type learning system includes: an original document image acquiring unit that acquires an image of a filled-in form as an original document image; an image reducing unit that reduces the original document image; a noise adding unit that adds a noise to the original document image, which has not been reduced by the image reducing unit or has been reduced by the image reducing unit, to generate a plurality of images for machine learning; a label adding unit that associates a form type of the original document image with the plurality of images for machine learning as a label; and a machine learning processing unit that performs machine learning of the classifier using the plurality of images for machine learning and the label as training data.