Skip to main content
Fig. 2 | IPSJ Transactions on Computer Vision and Applications

Fig. 2

From: Supervoxel-based segmentation of 3D imagery with optical flow integration for spatiotemporal processing

Fig. 2

Left: Suppose we want to calculate the affinity value between voxels A and B. C is the voxel with maximal mPb value that lies on the line segment \(\bar {AB}\). In the upper-left case, A and B belong to one region. In the lower-left case, A and B are in two different regions. But the intervening contour cue of UCM [1] gives the same affinity value in both cases, which is not very satisfactory. Obviously it would be better if we consider the direction \(\vec {t}\) of C’s mPb. Right: In our oriented intervening contour cue approach, when calculating affinity values, we always take the product of mPb(C) with the absolute value of the inner product \(\langle \vec {t},\vec {n}\rangle \), where \(\vec {n}\) is the unit direction vector of line segment \(\bar {AB}\). If A and B are on different sides of a boundary surface, \(|\langle \vec {t},\vec {n}\rangle |\) will be large, leading to small affinity value and vice versa

Back to article page