OCRchie User Interface Guide/ Advanced version

Brief User Interface Guide

Starting OCRchie

Start OCRchie with the name of your executable (in unmodified form, this is probably "orchie"). Make sure that your DISPLAY environment variable is set correctly and that you have display privileges. OCRchie will give an unhelpful "Bus Error" message if there is a problem with the display.

Opening an image

To open an image for OCR, select "Open" from the "File" menu. A small box displaying the contents of your directory will appear. First select Single zone if the document does not have multiple columns or images, or Multizone if you would like to zone this document. Double click on directory names. to change directory, or double click on a filename to open an image. If the image is not a 2-level (black and white) TIFF file, OCRchie may crash. If you selected Single zone, the image will be displayed in the upper window as long asi the "Display Image" option is enabled. If you selected Multizone the image will display in the zoning window. See the Zoning guide for information on zoning.

Marking Equations

Press the right mouse button at the beginning and end of each equation. The equation will be highlighted in yellow. You may delete equations by double clicking the middle mouse button within the equation. Marking an equation serves three purposes. First, it disables projection in component analysis. This eliminates inappropriate joining of integral limits, fractions, etc. Secondly it signals recognition to store each character as a separate word. Lastly its boundaries will be output in the Equation Bounds output s election underthe File menu.

Recognizing an image

Select "Recognize" from the "Tools" menu in order convert the image into text. The resulting text will be displayed in the bottom portion of the user interface main window and can be editted.

Learning

Select "Learn" from the "Tools" menu to learn the changes that you have made in the edit window. OCRchie will synchronize words upon learning, so make sure that the number of components in the word matches the number of characters in the translation. Components can be split or joined by double clicking the left mouse button and choosing "Horizontal Split" or "Join" from the popup menu.
You may also learn an individual component by double clicking on the component, choosing "Learn" and entering the translation. After you have completed learning you can test the learned characters by choosing "Recognize" from the "Tools" menu. Finally save the learned data to file to be read in at a later time.

Skew correcting an image

Select "Deskew" from the "Tools" menu in order deskew a tilted image. This routine is called automatically by "Recognize" if it is not turned off in the "Options" menu.

Editing recognized text

The recognized text may be freely modified in the bottom window of the user interface through the keyboard. Highlighted words are likely to have errors. To speed up editing, a user may start of the beginning and use the Tab key to jump the cursor through the document to each highlighted word. If displayed, the image will scroll to show the area around the highlighted word when Tab is pressed.

Saving recognized text

Select "Save ASCII" from the "File" menu to save the recognized text and any changes you have made to it. You will need to specify a filename for the output.

Select "Save WORDBOX" form the "File" menu to save in a format indicating the words and their positions on the page. You may output the full text or just the equations that have been marked. The format is:
(word) (upperleft point) (lowerright point) (upper right point) (lower left point) \n
An extra new line is inserted between lines of text.

Saving Learned Data

Select "Save Learned Characters" from the "File" menu to write out any characters OCRchie has learned during its current run. These data files can be read in during later runs much more quickly than the corresponding image and text files. On later runs you may select "Read Learned Characters" to retrieve the learned information.

Linking new files for learning

Closing a file

Selecting "Close" from the "File" menu will remove an image from the display and any of its recognized text from the edit window. Images should be closed before opening others.

Options

There are several options available to the OCRchie user

Warning Levels: the user can set the confidence levels below which words will be highlighted

Display Line Boundaries: When enabled, this will display what are believed to be the line boundaries on the image. This only works if the display is at 100%.

Display Bounding Boxes: When enabled, this option will draw rectangles around the connected components in an image. Also only works at 100% display size.

Spellcheck: This option will spellcheck the recognized text when selected and highlight mispelled words. It runs ispell on each word in the text.

Display Image: Toggles between displaying and not displaying the image. Displaying can take upwards of 15 seconds on large images, so it may be advisable to turn this off when it is not needed.

Deskew Options: "No Deskew" prevents the image from being deskewed. Since deskewing may take about 10 seconds, if images are not skewed by more than about half a degree, a user may want to select this option. "Bitmap rotate" uses a direct coordinate transformation to rotate the image. It is very accurate, but can be slow. "RLEMap rotate" is somewhat faster, but somewhat less accurate. This option is not recommended

Global Variables: The following global variables can be set to adjust the tolerances for recognition

	Noise Tolerance - Line breaks are detected if there are less than
	  NoiseToleance pixels on a line.
	MinLineSize - Lines that are less than this many pixels high will
	  be merged with the closest adjacent line.
	MinVertSeparation - Components separated by fewer than this many
	  pixels vertically will be merged.
	MinHorizSeparation - Components separated by fewer than this many
	  pixels vertically will be merged.
	ConfidenceThreshold - If a characters is recognized with a lower
	  confidence, an attempt will be made to split or join the character.
	JoinTolerance - Splits will only occur if the character is joined
	   by fewer than JoinTolerance pixels.
Options and global variable settings will return to their default value if OCRchie is restarted. To change the default values edit the file link_vars.tcl

Zoom in/Zoom out


The Zoom in and Zoom out commands increase and decrease the size of the displayed image respectively.

Refresh

Refresh, located in the Tools menu, redisplays the image.

Zoning guide

The Zoning display will appear if a document is openned in multizone mode. The user may either automatically or manually zone the page. Autozoning simply joins components horizontally or vertically by the selected number of pixels in order to create regions. You may select "Autozone" from the zoning menu to adjust these merging parameters and zone the page. Once zones are created you may adjust the zoning by performing "Autozone" with new parameters or manually deleting and adding zones. To delete a Zone - on a red button . To Create a Zone - drag and release. To activate a Zone - within region (not on buttons). The image will appear in the main window. Wordbox output will use the (x,y) coordinates within the large document. To Draw arrow to Zone - on a blue button. (This is purely cosmetic at this point. There are currently no output options for multiple zones.)