For the second project in ELEC539, our group worked on image morphing. Morphing involves the image processin
g techniques of cross-fading and warping to morph one image into a completely different image. We divided the
tasks into two categories, morphing
algorithms and automatic feature detection algorithms.
In morphing the most difficult task is the warping of one
image into another image. It is the stretching and pulling
of the images that makes the morphing effect so realistic.
The actual morphing of the image can
be accomplished either by using morph points or morph lines.
Morph points are the markers that you set up on the start image
and the end image. The morphing program then uses these
markers to calculate how the initial image should bend/warp
to match the shape of the final image. The second method uses
lines (edges) instead of individual points. Both methods
produce very realistic morphing effects.
One of the most time consuming tasks in morphing is selecting
the points or lines in the initial and final image so that the
metamorphosis is smooth and natural. An algorithm to automatically
select the morphing markers would save a great deal of time
and money. To attack this problem
the other half of the project focused on automatically extracting
the critical points and lines for morphing. To do this we first looked
at various edge detection methods. We then used these techniques
to compare the lines found in the initial and final images. This was
a difficult task for two reasons. The first problem is the difference
in contrast between the two images. This causes the edges of the
eyes in one image to be of different amplitude from the eyes in another
image, making it very hard to match the two pairs. Secondly, translation
differences between the images make it difficult to identify which edge
should be mapped to which other edge. The question is one of whether the
edge is an eyelid, an eyebrow, or a lip.
In summary, we created two separate algorithms for morphing one image into
another image. We also succeeded in developing an automatic edge
extraction algorithm for selecting the lines and points of the metamorphosis.
While the individual selection of the morph points still produces a
better transformation,
we feel that we made great strides toward an automatic morphing process
and that in the future such a program is possible.