Method for identifying a sequence of alphanumeric characters

• US 6,026,177 A
• Filed: 08/29/1995
• Issued: 02/15/2000
• Est. Priority Date: 08/29/1995
• Status: Expired due to Term

• Current Assignee: Asia Vision Technology Limited
• Original Assignee: Hong Kong University of Science and Technology
• Inventors: Kin, Wong Wing, Mong, John Lee Chung
• Primary Examiner(s): Johns, Andrew W.
• Application Number:
  US08/520,574
• Time in Patent Office: 1,631 Days
• Field of Search: 382/156, 382/157, 382/158, 382/159, 382/190, 382/187
• US Class Current: 382/156
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/147 Determination of region of interest
• CPC Class Codes: G06V 30/244 using graphical properties, e.g. alphabet type or font

Abstract

A character recognition system is described, in particular a system suitable for use in monitoring cargo container codes or vehicle number plates. An image of the code is first analyzed to extract potential characters. As part of this process, long horizontal and vertical line segments are filtered out. The extracted potential characters are then input to a two-level character recognition means. The first level comprises a neural network classifier that classifies a character into a smaller set of possible characters; and then the second level comprises another neural network classifier which identifies which character among the smaller set of possible characters the extracted character is.

Claims

1. A method of identifying a sequence of alphanumeric characters, comprising the steps of:

(a) producing a grey level image of said character sequence,

(b) locating potential characters from said grey level image including filtering said image with a long horizontal and long vertical line segments removal filter,

(c) normalizing the size and contrast of said extracted potential characters as grey level character images, and

(d) inputting said normalized potential characters as grey level character images to a character recognition means, wherein said character recognition means comprises two network classifiers, a first one of said network classifiers comprising a first hidden layer feed forward neural network which classifies the input pattern into a first set of categories, and the second one of said network classifiers comprising a second hidden layer feed forward neural network which recognizes the pattern from the first set of pattern categories.

2. A method of identifying characters as claimed in claim 1 wherein after filtering said image, successive horizontal and vertical segmentation techniques are used to segment the regions potentially containing characters.

3. A method of identifying characters as claimed in claim 2 wherein said segmentation techniques include removing unlikely potential characters.

4. A method of identifying characters as claimed in claim 2 wherein after segmentation the segmented regions are binarized and connected components are found by a four-way recursive component search.

5. A method of identifying characters as claimed in claim 2 including searching for lines of potential characters of similar heights and widths.

6. A method of identifying characters as claimed in claim 2 wherein said horizontal and vertical segmentation techniques include locating a local minimum in a horizontal or vertical projection followed by a vertical or horizontal local minimum, respectively, of the segmented regions to provide the horizontal and vertical segmentation points.

7. A method of identifying characters as claimed in claim 1 wherein characters are normalized to a size of 16×

16 pixels.

8. A method of identifying characters as claimed in claim 1 wherein after normalization the foreground and background colors are determined.

9. A method of identifying characters as claimed in claim 1 wherein said first network classifier comprises 100 hidden nodes and 36 output nodes.

10. A method of identifying characters as claimed in claim 1 wherein said second network classifier comprises 100 hidden nodes.