<?xml version="1.0" standalone="yes"?>
<Paper uid="P97-1038">
  <Title>An Alignment Method for Noisy Parallel Corpora based on Image Processing Techniques</Title>
  <Section position="4" start_page="301" end_page="302" type="relat">
    <SectionTitle>
4. Result and Discussion
</SectionTitle>
    <Paragraph position="0"> Figure 6 shows that the coverage and precision of the LTP estimate is not very high. That is to be expected since the translation is not literal and the mutual information estimate based on an outside source might not be relevant. Nevertheless, PlotAlign algorithms seem to be robust enough to produce reasonably high precision that can be seen from Figure 3. Figure 3(a) shows that a normalization and thresholding process based on one-to-one constraints does a good job of filtering out noise. Figure 3(b) shows that convolution-based filtering remove more noise according to the assumption of structure preserving constraint.</Paragraph>
    <Paragraph position="1"> Texture analysis does an even better job in noise suppression. Figure 7(a) and 7(b) show that signal-to-noise ratio (SNR) is greatly improved.</Paragraph>
    <Paragraph position="2"> The filtering based on Hough Transform, contrary to the other two filtering methods, prefers connection that is consistent with other connections globally. It does a pretty good job of identifying a long line segment. However, isolated, short segments, surrounded by deletions are likely to be missed out. Figure 8(b) shows that filtering based on HT missed out the short line segment appearing near the center of the dotplot shown in Figure 6(b).</Paragraph>
    <Paragraph position="3"> Nevertheless, this short segment presents most vividly in the result of textural filter, shown in Figure 7(b). By combining filters on all three levels of resolution, we gather as much evidence as possible for optimal result.</Paragraph>
  </Section>
class="xml-element"></Paper>