Digital Signal Processing Reference
In-Depth Information
Fig. 2.4
Simulation function block diagram of video object segmentation
noise. After join considering with signal space, the smallest eigenspace is too
sensitive to extract the objects that contains too many shadows. These shadows
usually are with a similar hue as the desired samples that we should take them
as parts of the objects. In order to correct representing the segmented objects, we
can take the advantage that the extracted color from the second large noise plane
is not so accurate as the smallest one. By inspecting the second noise space re-
lated to the second large eigenvalue, we find out that this plane can well describe
the silhouette of video object. Hence, the proposed color segmentation using com-
plete eigen-structure should result effective and efficient detection performances in
detecting the desired color objects. Figure
2.4
shows the procedures of applying
the proposed color eigen-structure segmentation method on some known or instant
samples of the desired color object such as skin, hair, or clothes.
2.2.4
Simulation Results
In order to verify the effectiveness of the proposed algorithms, we adopt four stan-
dard sequences, Mosaic, Ball, Akiyo, and News, which are shown in Fig.
2.5
in
simulations. For each image, we mark with a triangular sign on it to indicate the
desired color object we want to extract. Because we have not introduced any spa-
tial and temporal information, all the other similarly colored objects with the same
color sensation will also appear in the segmentation results. To clearly exhibit the
desired color in the segmented images, we only show the desired color and unde-
sired color objects by bright pixels with gray level
=
255 and dark pixels with gray
level
0, respectively. First, we only apply the PCT algorithm to extract the de-
sired color objects. From the signal and noise subspaces, the segmented objects are
varied diversely by different threshold setting. By using the threshold determination
=
