Segmentation of text, picture and lines of a document image

• US 5,335,290 A
• Filed: 04/06/1992
• Issued: 08/02/1994
• Est. Priority Date: 04/06/1992
• Status: Expired due to Term

• Current Assignee: Ricoh Company Limited
• Original Assignee: Ricoh Company Limited, Ricoh Corporation (Ricoh Company Limited)
• Inventors: Cullen, John F., Ejiri, Koichi
• Primary Examiner(s): Conso, Yon J.
• Application Number:
  US07/864,423
• Time in Patent Office: 848 Days
• Field of Search: 382/9, 382/44, 382/46, 382/56, 382/41, 358/261.1
• US Class Current: 382/176
• CPC Class Codes: G06T 9/005 Statistical coding, e.g. Huffman, run length coding
• CPC Class Codes: G06V 30/413 Classification of content, e.g. text, photographs or tables

Abstract

In a character recognition system, a method and apparatus for segmenting a document image into areas containing text and non-text. Document segmentation in the present invention is comprised generally of the steps of: providing a bit-mapped representation of the document image, extracting run lengths for each scanline from the bit-mapped representation of the document image; constructing rectangles from the run lengths; initially classifying each of the rectangles as either text or non-text; correcting for the skew in the rectangles; merging associated text into one or more text blocks; and logically ordering the text blocks.

Claims

1. In a character recognition system, a method for segmenting portions of a medium into text and non-text types, said method comprising the steps of:

a) compressing a bit mapped representation of said medium, said compressing said bit mapped representation of said medium including

a.i) providing said bit mapped representation of said medium to a compression means,

a.ii) compressing group of N scanlines of said bit mapped representation into a corresponding compressed scanline, and

a.iii) constructing a compressed representation of said medium from said compressed scanlines;

b) providing said compressed representation of said medium to a run length extraction and classification means, said compressed representation comprised of one or more scanlines;

c) extracting run lengths from each scanline of said compressed representation of said medium;

d) creating a run length record for each extracted run length, each run length record including a classification of the corresponding run length as short, medium or long based on it'"'"'s length;

e) constructing rectangles from said run length records, said rectangles representing a portion of said medium;

f) determining a skew of said rectangles;

g) correcting for skew of said rectangles;

h) classifying each of said rectangles as type image, vertical line, horizontal line or unknown; and

i) merging rectangles of type UNKNOWN into one or more text blocks.

2. In a character recognition system, a method for segmenting portions of a medium into text and non-text types, said method comprising the steps of:

a) providing a bit mapped representation of said medium to a compression means;

b) compressing groups of N scanlines of said bit mapped representation into a corresponding compressed scanline by performing the steps of;

b1) examining corresponding bytes of said N scanlines and assigning a first or second logical value to a corresponding bit of a corresponding byte of a temporary con, pressed scanline according to the rules;

assigning a first logical value if any corresponding bits of said corresponding bytes of said N scanlines has said first logical value;

assigning a second logical value if none of said bits of said corresponding bits of said corresponding bytes of said N scanlines has said first logical value;

b2) assigning all bits in a corresponding byte of a compressed scanline said first logical value if any bits in a corresponding byte in said temporary compressed scanline has said first logical value; and

b3) assigning all bits in a corresponding byte of a compressed scanline said second logical value if no bits in said corresponding byte in said temporary compressed scanline has said first logical value;

c) constructing a compressed representation of said medium from said compressed scanlines;

d) providing said compressed representation of said medium to a run length extraction and classification means, said representation comprised of one or more scanlines;

e) extracting run lengths from each scanline of said compressed representation of said medium;

f) creating a run length record for each extracted run length, each run length record including a classification of the corresponding run length as short, medium or long based on it'"'"'s length;

g) constructing rectangles from said run length records, said rectangles representing a portion of said medium;

h) determining a skew of said rectangles;

i) correcting for skew of said rectangles;

j) classifying each of said rectangles as type image, vertical line, horizontal line or unknown; and

k) merging rectangles of type UNKNOWN into one or more text blocks.

3. The method as recited in claim 2 wherein N is 4.

4. The method as recited in claim 2 wherein said step of extracting run lengths for each scanline of said representation of said medium further comprises the steps of:

a) identifying in a compressed scanline, a first byte that has said first logical value;

b) identifying a second byte having said second logical value in said compressed scanline wherein all the bytes between said first and second bytes have said first logical value;

c) storing a first address associated with said first byte and a second address associated with said second byte in a run length record associated with said compressed scanline.

5. The method as recited in claim 4 wherein said first logical value is a non-zero value and said second logical value is a zero value.

6. The method as recited in claim 5 wherein said non-zero value corresponds to a black pixel value and said zero value corresponds to a white pixel value.

7. The method as recited in claim 6 wherein said step of constructing rectangles from said run lengths is further comprised of the steps of:

a) providing a first set of run length records corresponding to a current compressed scanline to a rectangle construction means;

b) providing a second set of run length records corresponding to a preceding compressed scanline to said rectangle construction means;

for each run length in said first set of run length records

c) creating a new rectangle when a run length record is not related to any of said run length records in said second set of run length records;

d) assigning a run length to an existing rectangle when a relationship exists between a run length record in said first set of run length records and a run length record in said second set of run length records; and

e) updating a run length type counter for the run length type of said run length, said run length type counter associated with said rectangle.

8. The method as recited in claim 7 wherein said step of classifying each of said rectangles is further comprised of the step of providing a set of threshold values for said classification and classification is performed according to the following rules:

a) IF (Runlengths are all of Type Long ) AND (Rectangle Height is less than or equal to the Runlength Type Short threshold) THEN classify rectangle as type HORIZONTAL LINE;

b) IF (Runlengths are of Type Short) AND (Rectangle Height is greater than the Runlength Type Short threshold) THEN classify rectangle as type VERTICAL LINE;

c) IF ((Runlengths are Type Long) OR (Rectangle width is less than or equal to the Runlength type Long Threshold)) AND (Rectangle height is greater than an Image Height Rectangle Threshold);

THEN classify rectangle as type IMAGE; and

d) Classify all remaining rectangles as UNKNOWN.

9. An apparatus for segmenting portions of a medium into text and non-text types, said apparatus comprising:

a) scanline compression means for compressing N scanlines of a bit-mapped representation of a document image into one compressed scanline;

b) run length extraction means coupled to said compression means, said run length extraction means for extracting and classifying run lengths from each compressed scanline, and further for storing each of said run lengths in a run length storage means;

c) rectangle construction means coupled to said run length storage means, said rectangle extraction means for constructing and classifying rectangles and storing in a rectangle storage means;

e) skew correction means coupled to said rectangle storage means, said skew correction means for correcting a skew angle of said rectangles;

f) rectangle classification means coupled to said rectangle storage means, said rectangle classification means for assigning a classification to each of said rectangles as type image, vertical line, horizontal line or unknown;

g) merging means coupled to said rectangle storage means, said rectangle classification means for merging rectangles of the type unknown into text blocks and storing in said rectangle storage means; and

h) coordinate resolution means coupled to said rectangle storage means, said coordinate resolution means for resolving rectangle and block information in said rectangle storage means with real coordinate addresses of said bit-mapped representation of said document image.

10. The apparatus as recited in claim 9 is further comprised of block ordering means coupled to said rectangle storage means, said block ordering means for providing an ordering of text blocks.