Experiment results on extracting chicken area
The proposed ABLSE method was used to capture the real contour
of chickens. Here the condition of #1 images is traditional
warm-light, and the conditions of #2 images is white-light.
Experiments for two kinds of images were performed, one is
from the traditional warm-white light source and the other is from
the white light source. The observing images and segmentation
results are shown in Fig. 7. The upper row images in this figure
are for condition#1, and bottom row images are for condition#2.
In which, Fig. 7(a) and (h) are the background frames near to the
last phase in preheating. Fig. 7(b) and (i) are the first frames close
to commence of roasting. Fig. 7(c) and (j) are the binary foreground
images after using the background subtraction for the mask. Fig. 7
(d) and (k) are the #1 sub-image with 282-by-209 pixels and the
#2 sub-image with 347-by-251 pixels, respectively, in which its
maximal rectangle box is regard as a Mask for subsequent frame
segmentation when implementing the proposed segmentation
algorithm. Fig. 7(e) and (l) are the sub-images with a smaller rectangle
as an initial curve inside the target chicken because the
ABLSE always needs an initial curve. For convenience of comparison,
the initial curve is set to the same location in testing frames
of each video. Fig. 7(f) and (m) are final results by using the ABLSE
segmentation algorithm. One can observe visually that the contour
from #2 images is closer to the real target boundary than these for
#1 images.