视觉显著性特征约束下的骨架生长算法

Skeleton-growing algorithm based on visual saliency features

骨架特征在图像处理和计算机图形学等领域有着广泛的应用,而常规的骨架提取算法易受到噪声和物体自身变化的影响,使得提取的骨架难以进行后续的应用。本文提出联合离散曲线演化和弯曲度比率这两种视觉显著性特征约束下的骨架生长算法。在骨架生长过程中通过对判为结点的骨架点的邻域骨架点进行进一步的弯曲度比率约束,有效抑制了离散曲线演化约束骨架提取算法对于弯曲度较大的部位所产生的无法避免的冗余枝。通过调节保留的离散曲线演化点数以及弯曲度比率阈值,可获得多尺度的骨架。实验证明,在较大非刚体形变和轮廓噪声等干扰下,本文提出的算法仍能有效的抑制冗余骨架枝的产生,获得的骨架能够较好的表示图形中视觉重要部分。

The skeleton is a very useful shape descriptor and has been widely applied in image processing and com- puter graphics. However, shape skeleton extraction is usually highly affected by the boundary noise and large deformation of object, which cannot be used in further applications. To overcome this problem, a novel skeleton growing algorithm combining the strength of two complementary shape saliency measures, i.e. discrete curve evolution （DCE） and bending potential ratio （BPR）, is proposed. During the skeleton growth, the compact skeleton is directly obtained and the poten- tially redundant segments are pre-constrained under the control of the two introduced visual saliency measures. This strategy avoids the pruning stage naturally. Also the level-of-detail skeleton can be generated under the control of DCE and BPR. The experimental results on noisy and non-rigid deformed shapes demonstrate the valid of the proposed algo- rithm in the paper.