Method and system for single pass optical character recognition

A system for performing single pass optical character recognition (OCR), the system comprising:

• a network interface configured to receive an input image having machine printed characters;
• a fully convolutional neural network (FCN) engine having at least one processor;
• and
• at least one memory including instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
• preprocessing the input image having the machine printed characters;
• extracting two or more image features from the input image using a plurality of convolutional layers included in the FCN engine, wherein the two or more image features are extracted at each convolutional layer of the plurality of convolution layers included in the FCN engine;
• transforming feature maps from the plurality of convolutional layers to a size of the input image using a transposed convolution layer of the FCN engine;
• generating a multi-channel pixel-level probability output based on the transformed feature maps, wherein predictions in the multi-channel pixel-level probability output correspond to pixels of the input image in spatial dimensions;
• scaling the multi-channel pixel-level probability output to the size of the input image;
• generating word boxes for the machine printed characters based on the scaled multi-channel pixel-level probability output;
• determining each character of the machine printed characters in each of the word boxes based on the scaled multi-channel pixel-level probability output, wherein the determining each character comprises:
• multiplying aggregated character predictions for each of the word boxes by a character weighting index, and
• selecting each character based on a highest prediction from a result of the multiplying;
• and
• transmitting, for display, the machine printed characters determined from each of the word boxes.

A non-transitory computer-readable medium having stored thereon computer-readable instructions executable to perform single pass optical character recognition (OCR) utilizing a fully convolutional neural network (FCN) engine, the computer-readable instructions executable to perform operations which comprise:

• preprocessing an input image having machine printed characters;
• extracting two or more image features from the input image using a plurality of convolutional layers included in the FCN engine, wherein the two or more image features are extracted at each convolutional layer of the plurality of convolution layers included in the FCN engine;
• transforming feature maps from the plurality of convolutional layers to a size of the input image using a transposed convolution layer of the FCN engine;
• generating a multi-channel pixel-level probability output based on the transformed feature maps, wherein predictions in the multi-channel pixel-level probability output correspond to pixels of the input image in spatial dimensions;
• scaling the multi-channel pixel-level probability output to the size of the input image;
• generating word boxes for the machine printed characters based on the scaled multi-channel pixel-level probability output;
• determining each character of the machine printed characters in each of the word boxes based on the scaled multi-channel pixel-level probability output, wherein the determining each character comprises:
• multiplying aggregated character predictions for each of the word boxes by a character weighting index, and
• selecting each character based on a highest prediction from a result of the multiplying;
• and
• transmitting, for display, the machine printed characters determined from each of the word boxes.