The nal step is to produce stereo pairs at the end from
2D original images and nal obtained depth maps. There
are two ways of stereo generation. One is, consider original
2D image as left (or right) image and generate other image
from this original image. In this case, more artifacts will be
produced. One image is xed and all the warping process is
applied to other image, thus, artifacts will be present only
on the newer image and their dimension can be multiplied
by two.
The other method is to consider original 2D image as in-
termediate image and generate left and right views from this
image. The right and the left images will be half shifted or
warped toward the respective direction and all the artifacts
introduced by the processing will be part of both images but
halved visible. So, less artifacts are produced in this method.
We applied this technique [16] for nal stereo generation.
The rst row of g. 5 shows the original sharply focused
input image and bokeh image produced by our algorithm.
g. 5 second row shows depth from focus and motion which
is estimated from bokeh image. These two depths are com-
bined by weighted average which is shown in left of third
row and we get nal depth map.