基于对称轴的自然图像中物体部件检测

Symmetry-based object part detection in natural images

提出了1种基于对称轴的物体部件检测方法,将对称轴作为物体部件检测的种子。首先,基于学习几何与表象特征的相似度,将对称轴片段连成对称轴段,每个对称轴段被当作1个物体部件的中轴。然后通过定位每个对称轴段两边的物体部件边界检测得到物体部件。通过在BSDS-Parts数据集上的实验,证明了所提出的方法能够提高对物体部件检测的性能。

A method based on symmetry is proposed to detect object parts, in which symmetry is served as seeds. Firstly, symmetries are linked into segments based on the local similarity learned on geometric and appearance features. Each symmetry segment is considered to be the medial axis of an object part Then, driven from each symmetry segment, each object part is detected by locating the parallel part edges at the two sides of the symmetry segment Experimental results on the BSDS-Parts dataset and demonstrate the proposed method can improve the performance of object part detection.