Back to Aurora Vision Studio website

You are here: Start » Filter Reference » Computer Vision » Deep Learning » MergeCharactersIntoOneLine

MergeCharactersIntoOneLine

Advanced

Module: DL_OCR

Converts the output of the Deep Learning filter DL_ReadCharacters into a single text value, with the detected lines separated by a new line character.

Name Type Range Description
Input value inCharacters OcrResultArray Output of DL_ReadCharacters
Input value inMaxGap Real 0.0 - 10.0 Maximum gap between adjacent characters along the reading direction, denoted as fraction of 'A' char height
Input value inMaxShift Real 0.0 - 1.0 Maximum misalignment between adjacent characters perpendicular to the reading direction, denoted as fraction of 'A' char height
Input value inMargin Real 0.0 - 10.0 Additional margin added to result, denoted as fraction of 'A' char height
Input value inMinLength Integer 1 - 200 Minimal number of chars to create line
Input value inPattern GrammarRulesPattern Pattern used in Grammar rules filtering; a new line escape sequence is supported to match across detected lines
Input value inCandidates OcrCandidateArrayArray* Candidates - optional output of DL_ReadCharacters, Required when using grammar rules (when inPattern is not empty)
Input value inMinScore Real 0.0 - 1.0 Minimum score for filtering the line of text
Input value inLineDirection LineDirection Expected text orientation direction
Output value outText String All merged characters as a single text value, with the detected lines separated by a new line character
Output value outLine Rectangle2D Minimal box which covers all selected character boxes
Output value outScore Real Calculated score for the whole result
Output value outMapping Integer?Array Mapping between input characters and the output text, outMapping[i] is 0 if inCharacters[i] has been added to the result, NIL otherwise
Output value outProcessedCharacters OcrResultArray Contains the full set of character results after grammar rule processing. Maintains a one-to-one correspondence with inCharacters, preserving original positions but possibly with modified values, scores, or other attributes based on grammar rule application.

Description

This tool takes the text contained in OcrResultArray from the DL_ReadCharacters and merges it into a single text value returned in outText. Characters placed in the same line are joined with a space (exactly like MergeCharactersIntoLines with inFlatten = True), and the detected lines, stacked one above another, are joined with a new line character (\n). The whole result is therefore returned as one value together with a single bounding box outLine and a single outScore, instead of an array of lines.

Use MergeCharactersIntoLines when you need each line as a separate result; use this filter when the text should be read as one multi-line value.

Grammar Rules:

This feature can be used if we know the structure of the text we want to read. Define the inPattern string that you want to match against OCR results using regex syntax. This function uses more internal information inCandidates from Deep Learning filter DL_ReadCharacters to achieve the best matching to inPattern. Pattern is concatenation of pattern element. With this filter the grammar rules run once over the whole text, so a pattern may also span several lines (see the \n element below).

Note: To use grammar rules the inputs: inCharacters, inPattern, inCandidates are required


Pattern elements can be:
  • Individual character: one of character supported in DL_ReadCharacters
    • Escaping operational characters: Use a backslash to treat operational characters as normal: \\, \*, \?, \., \+, \-, \], \[, \), \(.
    • Whitespace Macro: \s represents a whitespace gap between words on the same line, space is also supported.
    • New line: \n matches a line break between two detected lines. Unlike \s (and space), which match a whitespace gap between words on the same line, \n matches only the transition to the next line, so a single pattern can span several lines. Example: \w+\n\d+ matches a word on one line directly followed by digits on the next line. Note: . (dot) does not match a line break.
  • Character class is a set of characters enclosed within square brackets []. It allows you to match any one character from the specified set. Example of usage:
    • List of characters: [abc] Matches any one of the characters a, b, or c.
    • Range: [a-z] Matches any one of the characters from a to z.
    • Mix of them: [a-zA-Z12] Matches any one of the characters from a to z, A to Z and 1,2.
    • Predefined character classes:
      • \d is equivalent to [0-9]
      • \w corresponds to [a-zA-Z0-9_]
      • . (dot) matches any single character on a line (\w plus special characters); it does not match a line break
      Note: Inside a character class, the following characters require escaping with a backslash \\, \-, \].
      Ex. [a.*|] is valid pattern which matches with the characters: a,.,*,|.
  • Chain is an extended string created by concatenating individual characters and character classes. Instance:
    • abc - matches text abc
    • [Aa]bc - matches texts: abc and Abc
    • \dabc - matches texts: 0abc, 1abc, ..., 9abc
  • Alternative is used to match one pattern or another. It's sequence of chains separated by pipe symbol | in round brackets (). Demonstration:
    • (abc|def) matches texts: abc and def
    • ([Aa]bc|\dabc) matches texts: abc, Abc, 0abc, 1abc, ..., 9abc

      • Note: Round round brackets are required. Ex. a|b
      Note: Nested brackets aren't supported. Ex. (a|(b|c))
  • Special operators can modify or repeat the preceding expression.
    • * (star): means zero or more occurrences of the preceding expression (in particular ".*" means any sequence), but tries to match as many characters as possible
    • + (plus): means one or more occurrences of the preceding element, maximizing the number of characters matched.
    • ? (question mark): means zero or one occurrence of the preceding element, with a preference for one.
    • *? (lazy star): means zero or more occurrences of the preceding expression, but tries to match as few characters as possible.
    • +? (lazy plus): means one or more occurrences, but minimizes the number of characters matched.
    Special operators can be added after the pattern element; Ex. [ABC]*, [0-9]?, (ABC|DEF)+
    Note: Special operators cannot be used inside alternative: Ex. (a*|b)

Depending on the inMaxGap and inMaxShift values, characters are grouped into a different number of lines, which then end up separated by a new line character in outText. See the image below, where increasing the inMaxGap results in one line of text, whereas a smaller value will return two separate lines: Different values of the inMaxGap and inMaxShift parameters result in a different number of lines.

The lines are sorted by the Y value before being joined with a new line character.

The tool can also be used to get rid of false characters by setting a different value of the inMinLength parameter, which filters out short groups of characters returned by the DL_ReadCharacters tool.

The inLineDirection parameter specifies the text orientation: Horizontal for left-to-right text, Vertical for top-to-bottom text. When set to Vertical, characters are grouped into columns instead of rows, and reading order follows the vertical axis. Effect of setting inLineDirection to vertical and horizontal.

Examples

Assume DL_ReadCharacters returned the following two-line label:

 PRD:2021/07/07 08:45
 BEST BEF0RE:2024/07/07

  1. Reading the whole label as a single value (empty inPattern). Characters on a line are joined with a space and the two lines with a new line character:

   inPattern= (empty)
   outText= PRD:2021/07/07 08:45\nBEST BEF0RE:2024/07/07

  2. Matching across the line break with the \n element - the time at the end of the first line followed by the beginning of the second line:

   inPattern= \d\d:\d\d\nBEST
   outText= 08:45\nBEST

  3. Matching a value that continues on the next line, e.g. a "best before" date split over two lines:

   inPattern= BEF0RE:\d\d\d\d/\d\d/\d\d
   outText= BEF0RE:2024/07/07

Remarks

This article concerns the functionalities related to another product: Deep Learning Add-on.

Errors

This filter can throw an exception to report error. Read how to deal with errors in Error Handling.

List of possible exceptions:

Error type Description
DomainError If you want to use grammar rules, please add inCandidates from filter DL_ReadCharacters.outCandidates and set True value on DL_ReadCharacters.inCalculateCandidates.
DomainError inCandidates and inCharacters must have the same size.

Complexity Level

This filter is available on Advanced Complexity Level.

See Also

  • DL_ReadCharacters – Performs optical character recognition using a pretrained deep learning model.