Digital Signal Processing Reference
In-Depth Information
IFEN is designed to effectively extract a feature subset from an original raw data-
set by using the computational principles inspired by the immune system [10]. Our
proposed method applies IFEN to extract an optimal foreground/background pair set
from the original sample set, then use the Robust Matting method on this "best" and
small set of foreground/background pairs directly to gain the resulting alpha matting.
The size of the searching space for the Robust Matting method is decreased greatly
and the quality of the candidate foreground/background pairs are improved a lot, so
that both the accuracy and the speed of matting can be promoted.
Next, in Section2, we introduce the basic framework of our proposed matting
solution with IFEN; in section3, the experimental results are presented. Finally,
conclusions are put forward in Section4.
2
Image Matting Based on IFEN
The procedure of the proposed matting method based on IFEN is shown in Fig. 1. The
first step of matting is to provide a trimap, which segments the input image into three
regions: definitely foreground Fg, definitely background Bg and unknown Cz. The
second step is to collect the original sample set from boundary of foreground and
background. Then, the "best" set of foreground and background pairs are extracted
from the original sample set by IFEN. Next, for each pixel
I
in the unknown area,
implement matting using the Robust Matting on the optimal foreground/background
pair set, resulting in the alpha matte of the input image as a set of estimated
*
z
α
. Final-
ly, in order to prevent discontinuities in the resulting matte, an additional step, matte
smoothing, is used to ensure the local smoothness of the final alpha values, while
maintaining its distinct features.
Fig. 1.
Flowcharts of IFEN-based matting
