Neural network-based diacritical marker recognition system and method

• US 5,373,566 A
• Filed: 12/24/1992
• Issued: 12/13/1994
• Est. Priority Date: 12/24/1992
• Status: Expired due to Fees

• Current Assignee: Info Enterprises, Inc., Motorola, Inc. (Motorola Solutions, Inc.)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Murdock, Michael C.
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): Fox, David
• Application Number:
  US07/996,440
• Time in Patent Office: 719 Days
• Field of Search: 382/9, 382/13, 382/14, 382/15, 382/36, 382/48, 382/58, 382/65
• US Class Current: 382/156
• CPC Class Codes: G06V 30/248 involving plural approaches, e.g. verification by template match; Resolving confusion among similar patterns, e.g. "O" versus "Q" G06V30/242 takes precedence

Abstract

A diacritical marker recognition system and method recognizes diacritical markers in a character image based upon an analysis by a neural network of the portion of the character image most likely to contain a diacritical marker. Once the neural network determines that a diacritical marker most likely exists in the character image, the system determines by using heuristics whether a diacritical marker exists or whether the character image appears to contain a diacritical marker which is actually a regular character.

Claims

1. A diacritical marker recognition system which is connectable to receive a raster image and bounding box information, said bounding box information specifying locations and dimensions of a rectangle surrounding characters in said raster image, said diacritical marker recognition system comprising:

extraction means for extracting a plurality of character images from said raster image based on said bounding box information;

subsampling means for subsampling an upper and lower region of each of said extracted character images, the subsampling in the upper region occurring at a different rate than the subsampling in the lower region;

neural network means for determining a probability of whether a diacritical marker may exist in each of said subsampled character images; and

controller means for determining whether a diacritical marker exists in one of said character images based on said determining by said neural network means and heuristics.

2. A diacritical marker recognition system as recited in claim 1, wherein said extraction means includes normalization means for normalizing each of said character images to a standard character image size.

3. A diacritical marker recognition system as recited in claim 1, wherein each of said character images includes a plurality of pixels, and wherein said subsampling means includes a second neural array means for subsampling said pixels in upper and lower regions of each of said normalized character images.

4. A diacritical marker recognition system which is connectable to receive a raster image and bounding box information, said bounding box information specifying locations and dimensions of a rectangle surrounding characters in said raster image, said diacritical marker recognition system comprising:

extraction means for extracting a plurality of character images from said raster image based on said bounding box information;

first neural array means for subsampling an upper and lower region of each of said extracted character images, the subsampling in the upper region occurring at a different rate than the subsampling in the lower region;

second neural array means for determining a probability of whether a diacritical marker may exist in each of said subsampled character images; and

controller means for determining whether a diacritical marker exists in said character image based on said determining by said second neural array means and heuristics.

5. A diacritical marker recognition method executed on a computer as part of a computer program for identifying if any character in a raster image contains a diacritical marker, said computer connectable to receive a raster image and bounding box information, said bounding box information specifying locations and dimensions of a rectangle surrounding each character in said raster image, said method comprising the steps of:

(a) extracting a plurality of character images from said raster image based on said bounding box information;

(b) subsampling an upper and lower region each of said extracted character images, the subsampling in the upper region occurring at a different rate than the subsampling in the lower region;

(c) identifying if any of said subsampled character images has a diacritical marker by using a neural network; and

(d) outputting said diacritical marker if said diacritical marker is identified in step (c).

6. A diacritical marker recognition method as recited in claim 5, wherein step (a) comprises the step of:

(a1) normalizing each of said character images to a standard character image size.

7. A diacritical marker recognition method as recited in claim 5, wherein step (b) comprises the step of:

(b1) subsampling an upper and lower region of each of said extracted character images.

8. A diacritical marker recognition method as recited in claim 5, wherein step (b) comprises the step of:

(b1) subsampling an upper and lower region of each of said extracted character images using neurons.

9. A diacritical marker recognition method as recited in claim 5, wherein step (b) comprises the step of:

(b1) subsampling an upper and lower region of each of said extracted character images using fixed-weight, binary threshhold neurons.

10. A diacritical marker recognition method as recited in claim 5, wherein step (c) comprises the step of:

(c1) identifying if any of said subsampled character images has a diacritical marker by using said neural network and heuristics.

11. A diacritical marker recognition method executed on a computer as part of a computer program for identifying if any character in a raster image contains a diacritical marker, said computer connectable to receive a raster image and bounding box information, said bounding box information specifying locations and dimensions of a rectangle surrounding each character in said raster image, said method comprising the steps of:

(a) extracting a plurality of character images from said raster image based on said bounding box information;

(b) normalizing each of said character images to a standard character image size;

(c) using a first neuron array to subsample an upper region of each of said normalized character images;

(d) using a second neuron array to subsample an lower region of each of said normalized character images, the subsampling in the upper region occurring at a different rate than the subsampling in the lower region;

(e) identifying if any of said subsampled character images from said first and second neuron arrays has a diacritical marker by using a neural network and heuristics; and

(f) outputting said diacritical marker if said diacritical marker is identified in step (e).