Texture Characterization for Outliers Rejection
a) Underwater image sequences. Dawnload area.
b) Textural operators analysis and threshold detection.
- This textural based approach was tested with fifteen underwater image sequences.
Thes images were selected because they are a good representation of the conditions
found in underwater environment, such as: blurring, lack of well-defined contours,
bad visibility, low contrast, scattering effects, non-uniform illumination,
lighting artifacts generated by the waves, etc. Moreover, different scenarios
were selected: rocked seafloor, sand and algae seafloor, and also some man-made
objects like an old submerged chain, moving fishes, etc. For every image sequence,
a set of point pairs was obtained considering a search window of 61 x 61.
Figures on the left side show the tested underwater sequences and their obtained
correspondences through region-based correlation.
- In the middle we can see the distribution of similarity measure between
two characterization vectors (point-mathcing).
- On the right side, the best textural operator: L5S5 energy filter with Absolute
mean was applied to each sequence for outliers rejection.
c) Applying texture operators.
L5S5 energy filter with Absolute mean was applied to different sequences ,
25 x 25 characterization window subsampled every two pixels were used