基于中心-对角对比度的显著性检测

Salient Detection Based on Corner-Surround Contrast

针对图像显著性检测中广泛运用的中心-四周对比度方法存在的缺陷,提出了中心-对角对比度(corner-surround contrast,CSC)算法,实现从背景中有效提取显著区域;中心-四周对比度方法难以保证分割的准确性,容易导致错误的显著检测结果,并且仅使用中心-四周对比度对目标物体进行显著性编码并不十分有效;CSC算法在提取中心和周边区域差异性的同时,结合它们之间位置的相对性,并设计了一个多核信息融合模型,以不同权重融合多种对比度方法以产生最优显著图,最后用图分割算法来进行二元分割,获得准确的显著图;实验结果表明CSC算法能够有效捕捉显著物体的局部差异,提高显著性物体定位及分割精度的性能,减少噪声的影响,从背景中分离出精确的显著性区域,获得更优的显著性检测结果。

To overcome the disadvantanges of center-surround contrast which is widely used for visual saliency detection, a novel con trast computational scheme, namely corner-surrond contrast, is presented to accurately detect salient regions from background. The center -surround contrast may involve the inaccurate segmentation, and even results in incorrect detection results. Only using center-surround contrast is not efficient to encode object saliency. The corner-surroud contrast not only considers the appearance difference between center and corner regions, but also takes into account their relative location. Then a kernel-based fusing framework is designed to produce the saliency map using a series of contrast measurements, and the final binary segmentation is achieved using graph cut algorithm. The experiments demonstrate that CSC can capture local dissimilarity of salient objects, and improve segmentation accuracy and saliency localization. Further- more, CSC can generate precise salient segmentation with less noises from the backgrounds, and gains significantly in terms of saliency detection.