30.11.2012 Views

Thesis (PDF) - Signal & Image Processing Lab

Thesis (PDF) - Signal & Image Processing Lab

Thesis (PDF) - Signal & Image Processing Lab

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Chapter 4<br />

The “trench” problem and the<br />

proposed solutions<br />

The BTV-domain morphological operators (erosion, opening etc.) proposed by Keshet<br />

in [16] are self-dual. This means that those operators treat bright and dark “objects”<br />

similarly. However, when applying those operators, a “trench” problem may arise.<br />

This chapter studies this problem, and proposes solutions for it.<br />

The BTV-domain morphological filtering approach involves three steps. First, the<br />

BTV transform is performed on the input image, using the TD-tree representation<br />

of the image. Then the transformed image is simplified using morphological filtering,<br />

based on the partial ordering of images of alternating sequences, defined in Section<br />

2.5.4. Finally output image is constructed from the processed image by an inverse<br />

BTV transform, using Equation (2.21).<br />

Figs. 4.1 and 4.3 show the output of the erosion ˆεB on two test images. Figs. 4.2<br />

and 4.4 show the corresponding opening ˆγB of the same two test images.<br />

Notice that the results of self-dual erosion and opening (Figs. 4.1-4.2) on the<br />

synthetic image are as expected. The corresponding erosion has the effect of shrinking<br />

all the elements of the image, regardless to their contrast. The opening rounds the<br />

corners that SE can not get into in all the elements of the image, bright and dark.<br />

However, the results on the natural image are very strange; the output shows<br />

many “trenches” that do not appear to have any justification to exist.<br />

These are typical results. When the number of gray levels is small, and the image<br />

is not very complex, the BTV-based erosion returns useful results. However, when<br />

44

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!