Abstract: In contour grouping computational models, the extraction of group-meta set has an important influence on the grouping result. In this paper, a group-meta set extraction algorithm based on global motion contrast is presented for the confusion of target and non-target edge fragments in complex scenes. It introduces similarity metrics for two edge segments based on motion features. Most of non-target edge fragments are removed from the original group-meta set on the basis of the definition for global motion contrast. And a group-meta set consisting of more target edge fragments and few non-target edge fragments is extracted. The method is tested on Moseg_dataset and it is compared with the methods based on edge detection and boundary detection. And the experiment results show substantial improvements in the scale and effectiveness for the extraction of group-meta set by this method. Further, it conducts groping experiments by the same grouping method for evaluation and comparison. Grouping results show that this method significantly outperforms previous algorithms.

Key words: extraction of group-meta, motion features, contour grouping, global motion contrast

摘要： 在轮廓编组计算模型中，编组元的提取对于轮廓编组结果具有重要的影响。针对复杂场景中目标轮廓易与非目标边缘混淆的问题，提出了一种基于全局运动对比度的编组元提取算法。提出了基于边缘片段的运动相似度度量方法，并通过相似度定义了场景中的全局运动对比度，以此对非目标边缘片段进行抑制，从而提取出更为有效的目标轮廓边缘片段构成编组元集合。在Moseg_dataset数据集上的实验结果证明，提出的全局运动对比度对于非目标边缘片段具有良好的鉴别能力，相比较目前轮廓编组计算模型中基于边缘检测和轮廓检测的编组元提取算法，该算法显著降低了编组元集合的规模，提高了编组元集合的有效性。在相同的轮廓编组算法中，该算法提取的编组元集合能取得更优的编组结果。

关键词: 编组元提取, 运动特征, 轮廓编组, 全局运动对比度