BACKGROUND FIELD
[0001] In general the subject under discussion communicates about identifying and removing
10 the skew from the text documents introduced during the scanning process.
DISCUSSION OF RELATED FIELD
[0002] Automatic de-skewing of scanned text documents, with or without embedded images,
15 are in demand for making the scanning process faster and human interference free.
[0003] Present document describes a novel and simple technique for performing this task.
The technique uses scanner output for processing, but could be utilized when the picture of a
document is captured with the help of a camera provided the focus is on the document.
20
DETAILS OF THE PROJECT
(00041 The said technique is able to de-skew the text documents with or without images
(comprising of m levels of shades; where m=2", n representing number of bits per pixel)
25 irrespective of location of the image in a text document.
[0005] The technique does not require any a priori information about the angle of skew in the
document.
30 METHOD
[0006]. The step wise procedure for skew estimation, detection and correction is discussed in
the present section.
3 5 [0007] 1. Scan a handwritten or computerized text document; refer Figure 1 (skewed
document). The scanned image, in Red-Green-Blue format, will be first converted to
greyscale; that is, image having only shades of gray. Each pixel in the grey image will be
depicted by 8 or more bits, depending on the image capturing device capabilities.
[OOOS] 2. A non-linear filter is then applied to reduce the noise artefacts from the image.
40 [0009] 3. Convert the image obtained in Step 2 to binary image.
[OOlO] 4. Complement operation is performed on the image, refer Figure 2.
[0011] 5. Define the text area of the image; shown in Figure 3 (we call it.BBi).
[0012] 6. Crop the image to BBi and denote the new image by I,, refer Figure 4.
[00131 7. Find the probabilistic distribution of white and black pixels.
45 [0014] 8. Define a range, %~[-cp to cp], within this range our algorithm will search for the
de-skewing angle. Smaller range will lead to less computation time but it may not lead to
exact de-skewing of image (in case it is skewed more than the range); higher range will
produce de-skewed document irrespective of its skew angle but will be computationally
expensive.
50 .[0015] 9. Define a'de-skew tolerance range, eO, of rotation and rotate the image by this
small non-zero user defined angle, this can also be predefined in the system, with the help
of following equation
where, (x, y) depicts the location of the .pixel to be counter clockwise rotated and (x', y')
is the rotated position of the respective pixel. .
[0016] 10. After rotating the image by O0 in clockwise (one can start with anticlockwise
direction as well, only fact to keep in mind is to scan the complete range, R,) direction, a
new image is generated with different size and background.
[0017] 11. Follow Steps 5 and 6 to extract text area in the rotated image and denoted this
new image by 1'.
[0018] .12. Calculate new probability distribution of white pixels in each row of image 1'.
[0019] 13. Rotate the image I' by e0 and repeat the procedure for all the angles defined
within the range, %.
[0020] 14. Among all these images, the image having the highest probability of white
pixels in each row, will be the de-skewed image I", refer Figure 5.'
[0021] 15. The angle corresponding to I", which is denoted by eFina=l k x e0 (where k
is the number of times Step 13 is repeated before fulfilling the condition stated in Step
14), will be the skew angle.
[0022] 16. Rotate the original image with angle eFinatlo generate the de-skewed image
(refer Figure 6).
CONCLUSION
75
[0023] The disclosure of the discussed methodblogy makes it possible to de-skew scanned
documents with varying angles. It is contemplated that few steps may be omitted or
performed parallel with other steps wherever required.
80 [0024] Although the process is discussed for calculating the angle of skew and its
' rectification for text only English language documents but it has capability for an extension
for any'other language and typed and handwritten documents with images etcetera embedded
within the document as well; e.g., refer Figures 7, 8, 9 and 10.
85 BRIEF DESCRIPTION OF DRAWINGS
[0025] Figure 1 : Original scanned image that is skewed at a random angle, while scanning.
[0026] Figure 2: Complemented skewed image corresponding to image obtained in Figure 1.
90
. [0027] Figure 3: Relevant text data of the scanned image show within a box with its
boundaries.
[0028] Figure 4: Cropped image 1'; the image is confined to the boundary region and size of
95 the figure is reduced.
[0029] Figure 5: De-skewed image obtained after the removal of skewness from the cropped
image that is obtained in Figure 4 following the steps mentioned in the article.
100 , [0030] Figure 6: Resultant de-skewed image of image shown in Figure 1.
[0031] Figure 7: Skewed document image containing bothtext and images.
[0032] Figure 8: Resultant de-skewed image of the scanned document shown in figure 7.
105
[0033] Figure 9: Handwritten skewed image of a document in Sanskrit language.
[0034] Figure 10: Resulting de-skewed image corresponding to Figure 9.

We claim:
1. An innovation to automatically detect and correct skew angle of a document that
is introduced during scanning the document, with the help of a scanner or an
115 image capturing device.
2. The method is capable of detecting and correcting' skew from text documents,
image within a box, bar codes, and any document with the combination of these.
3. The method described in the method section of the current document.