document_image_utils 0.1.24.4

Toolkit for document image processing

Contextualization

In progress toolkit for document image pre processing.

Aimed for images to be OCRed.

Main Available methods

• Auto rotate image

  Uses left margin of a document to calculate the angle of rotation present, and correct it accordingly.

  Can be given the rotation direction (clocwise or counter_clockwise), or in auto mode tries to determine the side to which the document is tilted (can be none, in which case image won't be rotated).

• Calculate rotation direction

  Calculates rotation direction of an image by finding the biggest sets of the first black pixels appearances (with outliers removed) in the image for each direction: clockwise, counter_clockwise and none.

  For none direction, the set is created based on pixels with same 'x' coordinate that with less than a 5% height difference, relative to the image's height.

• Binarize document

  Normal binarization with otsu tresholding and fastNlMeansDenoising.

  Fax binarization, following the image magick command: convert "image" -colorspace Gray ( +clone -blur 15,15 ) -compose Divide_Src -composite -level 10%,90%,0.2

• Split document into columns

  Analyzes document image pixel color frequency and split document image into columns.

• Auto crop document

  Analyzes document image pixel color frequency and cut document margins, aiming mostly to remove possible folds in the corners.

• Identify document images Identify document images in image, using algorithm available in leptonica's repository that finds potential image masks.

• Get document delimiters Get document delimiters, using image transformations.

• Segment document Segments document image into header, body and footer, using delimiters. Only the body is always guaranteed to have a value.

Bash commands:

• binarize : binarize document image.

• rotate_document : rotate document image.

• split_columns : split document into column images.

• d_auto_crop : auto crop document image.