Semantic Object Segmentation - Video Segmentation and Its Applications - page 76

Fig. 3.11 Examples of object segmentation results by LDA and SLDA. The images are from the
MSRC data set [ 2 ]. The first row shows example images. The second row uses manual segmenta-
tion and labeling as ground truth. The third row is the LDA result and the fourth row is the SLDA
result. Under the same labeling approach, image patches marked in the same color are in one ob-
ject cluster, but the meaning of colors changes across different labeling methods. The results are
from [ 11 ]

Fig. 3.12 Object segmentation from a video sequence. The first column shows two frames in the
video sequence. In the second column, the patches in the two frames as are labeled as different
object classes using LDA. The third column plots the object class labels using SLDA. In the fourth
column, tigers are segmented out by choosing all the patches of the class marked by red color. The
results are from [ 11 ]

3.4.3 Other Topic Models of Including Spatial Information

Some other topic models were also proposed to include spatial information. Russell
et al. [ 51 ] first obtained multiple segmentations of each image at different scales
using normalized cut [ 57 ] and then treated each segment instead of an image as a