Thesis (PDF) - Signal & Image Processing Lab
Thesis (PDF) - Signal & Image Processing Lab
Thesis (PDF) - Signal & Image Processing Lab
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
6.4. APPLICATION EXAMPLES 101<br />
Figure 6.28: The structuring element used in opening by reconstruction.<br />
6.4.2 Noise filtering for text OCR<br />
Another example of pre-processing for OCR can be seen in Fig. 6.37(a). The figure<br />
shows text taken from a National Geographic journal. This text is printed on non-<br />
uniform background, and worse, part of the text is white and part of the text is black.<br />
In addition the picture is corrupted by added random white and black pixels. If we<br />
want to use this image for OCR, the image must be cleaned. We used an EWT-<br />
based opening filter for this purpose. As can be seen in Fig. 6.37(b), both white<br />
and black text was cleaned in the same way. Both white and black noise pixels were<br />
removed, and the letters were almost undamaged. Although we didn’t run an actual<br />
OCR, we believe that the filtered image would have a higher success rate in OCR.<br />
For comparison, we filtered the image with the median filter and opening filter based<br />
on shape-tree. It can be seen in Fig. 6.38 that the median filter leaves some noisy<br />
pixels, and the shape-tree opening is almost identical to the EWT opening.