基于颜色、轮廓特征融合的显著性检测

Saliency detection based on the fusion of the color feature and contour feature

针对复杂场景中的显著性检测中所存在的困难,考虑到目标在纹理特征上的显著性,创新地采用颜色和轮廓特征融合的方法对传统显著性模型的输入进行改进,分别采用HSV色彩空间的颜色特征及基于Sobel算子的轮廓特征输入至Itti模型中对场景中的显著性进行检测,在复杂场景中直观及量化实验结果证明,该方法在较复杂的环境里能够较为准确的提取出显著性目标区域,同典型的Itti模型相比,能够有效提升显著性检测的准确性。

Aiming to solve the difficulties in saliency detection for complex scenes and considering the pop-out of the texture of the interest objects,in this paper we improve the input of the classical Itti model by using the novel color and contour features.The HSV color space feature and the contour feature based on Sobel operator are input into the Itti saliency extraction model to detect the saliency.The qualitative and quantitative experimental results show that the proposed method can extract the saliency target region more accurately in complex scenes and improve the accuracy of the saliency detection compared with the typical Itti model.