Whole word, phrase or number reading

• US 4,764,973 A
• Filed: 05/28/1986
• Issued: 08/16/1988
• Est. Priority Date: 05/28/1986
• Status: Expired due to Fees

• Current Assignee: The United States of America As Represented By The Secretary of Agriculture
• Original Assignee: United States Of America As Represented By The Secretary Of The Air Force
• Inventors: O'Hair, Mark A.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Skinner, A. Anne
• Application Number:
  US06/867,642
• Time in Patent Office: 811 Days
• Field of Search: 382/15, 382/14, 382/30, 382/43, 382/34, 382/48
• US Class Current: 382/229
• CPC Class Codes: G06V 10/431 Frequency domain transformation; Autocorrelation

Abstract

The image of a word is taken and the two-dimensional discrete Fourier transform of the image is computed. The transformed image is filtered to the first three harmonics, with both real and imaginary components. These components then make up a total of 49 unique vectors which defines a 49 orthogonal vector space. The vector space is normalized to unity and each image of a word or phrase defines a point within this 49 orthogonal, hypersphere. The same process is done to the image for the Fourier components, where there is only 25 unique vector components. Similar looking words cluster in the hypersphere and the smaller distance from one point to another defines the probability of incorrectly recognizing a word. In a study for the case of two through eleven letters in a word using both 49 and 25 vector space calculations, the results show two through eleven words are recognizable using 49 vector space and possibly the 25 vector space. The 25 vector space shows problems with symmetry (dyslexia) in many of the incorrectly recognized words, which was never the case for the 49 vector space. A conclusion is that people with dyslexia might use a different process to recognize words and by using the real and imaginary components, whole word recognition is possible.

Claims

1. The method of recognizing text, with a system having representative digital images stored in memory, comprising the steps:

(a) inputting an image representing text;

(b) digitizing the image to form a digital picture thereof;

(c) storing the image in memory;

(d) breaking the image into blocks of data, wherein the blocks represent entire words, phrases, or numbers;

(e) placing a window around the first individual block;

(f) performing a two-dimensional discrete Fourier Transform (2DDFT) of the image within the window;

(g) filtering to the first three harmonics, with both real and imaginary components, these components then making up a total of 49 unique vectors which define a 49 orthogonal vector space;

(h) energy normalizing to unity the 49 unique vectors, which will eliminate the effects due to image brightness;

(i) searching a library of known 49 orthogonal vectors and finding the closest match by euclidian distance;

(j) after finding a closest match, recognizing the image within the window as a particular word, phrase, or number, and storing the result in memory;

(k) placing a window around the next individual block and repeating steps (f) through (j) until all blocks within the image are recognized; and

(m) outputting the resulting phrases, words, or numbers as text;

and wherein symmetry properties are used so that the image for each block is recognized similar to a person having dyslexia.

2. The method according to claim 1, wherein the phase components of the 2DDFT are not used and only the unique 25 orthogonal vectors from the magnitude components are used, except for the exclusion of the phase components, steps (a) through (m) are repeated using the 25 vector space wherein the image recognized by the computer in step (i) will be similar to a person having dyslexia.