High accuracy optical character recognition using neural networks with centroid dithering
First Claim
Patent Images
1. A computer-implemented process of recognizing a pattern in an image among a set of known templates, the process comprising:
- a) training a neural network using said set of known templates;
b) scanning said image;
c) detecting said pattern by segmenting said scanned image into a detected pattern comprising a plurality of pixels, each such pixel having a value;
d) preprocessing said detected pattern by;
i) determining a minimum of said values of said pixels;
ii) subtracting the minimum from said values of said pixels;
producing thereby a corrected value for each such pixel; and
iii) filtering said corrected pixel values by selectively assigning a predetermined filtered pixel value to a subset of said pixels responsive to said corrected values of said pixels in said subset not exceeding a threshold value; and
e) recognizing said preprocessed detected pattern as corresponding to one of said known templates by applying said preprocessed detected pattern to said trained neural network.
3 Assignments
0 Petitions
Accused Products
Abstract
Pattern recognition, for instance optical character recognition, is achieved by training a neural network, scanning an image, segmenting the image to detect a pattern, preprocessing the detected pattern, and applying the preprocessed detected pattern to the trained neural network. The preprocessing includes determining a centroid of the pattern and centrally positioning the centroid in a frame containing the pattern. The training of the neural network includes randomly displacing template patterns within frames before applying the template patterns to the neural network.
38 Citations
3 Claims
-
1. A computer-implemented process of recognizing a pattern in an image among a set of known templates, the process comprising:
-
a) training a neural network using said set of known templates; b) scanning said image; c) detecting said pattern by segmenting said scanned image into a detected pattern comprising a plurality of pixels, each such pixel having a value; d) preprocessing said detected pattern by; i) determining a minimum of said values of said pixels; ii) subtracting the minimum from said values of said pixels;
producing thereby a corrected value for each such pixel; andiii) filtering said corrected pixel values by selectively assigning a predetermined filtered pixel value to a subset of said pixels responsive to said corrected values of said pixels in said subset not exceeding a threshold value; and e) recognizing said preprocessed detected pattern as corresponding to one of said known templates by applying said preprocessed detected pattern to said trained neural network.
-
-
2. A computer-implemented process of recognizing a pattern in an image among a set of known templates, the process comprising:
-
a) training a neural network using said set of known templates; b) scanning said image; c) detecting said pattern by segmenting said scanned image into a detected pattern represented as a map of pixels, each having a multi-bit pixel value; d) preprocessing said detected pattern by; i) determining complements of each of said pixel values; ii) selecting as a noise value a minimum of said complements; iii) subtracting from each of said complements said noise value; and iv) selectively replacing said complements with a baseline value in response to said complements not exceeding a predetermined threshold value; and e) recognizing said preprocessed detected pattern as corresponding to one of said known templates by applying said preprocessed detected pattern to said trained neural network.
-
-
3. Apparatus for recognizing a pattern in an image among a set of known templates, the apparatus comprising:
-
scanning means for scanning said image; a preprocessor adapted to accept as input an image signal produced by said scanning means and producing therefrom a preprocessed pattern signal, said preprocessed pattern signal including a plural number of multibit pixel values, said preprocessor including an inverter for determining complements of said pixel values, a noise filter for subtracting from said complements a minimum value thereof to obtain corrected complements, said noise filter further selectively replacing said corrected complements with a baseline value in response to said corrected complements not exceeding a predetermined threshold value; and a neural network adapted to accept as input said preprocessed pattern signal and to produce therefrom an output signal indicative of said pattern corresponding to one of said templates.
-
Specification