基于Log-Gabor滤波与显著图融合优化的3D显著性检测

3D Image Saliency Detection Based on Log-Gabor Filtering and Saliency Map Fusion Optimization

提出了一种基于Log-Gabor滤波的纹理和深度图融合优化的立体图像显著性检测模型,利用平面图像的显著性结合纹理与深度特征检测立体图像的显著性。通过改进的基于图的流行排序算法计算左视点的显著图;提取左视点图像的纹理特征图以及立体图像的深度特征图,利用Log-Gabor滤波器分别计算深度显著图和纹理显著图;再利用线性加权融合方法将上述3个显著图融合为立体(3D)显著图;最后利用中心偏爱和视觉敏锐度增强3D显著图。实验利用公开的眼动跟踪数据库进行测试,结果表明,所提算法具有很好的检测效果,优于文献报道的其他3D显著性模型。

A saliency detection model is proposed based on Log-Gabor filtering and saliency map fusion optimization of stereoscopic images,in which the image saliency is detected by the planar image saliency combined with the texture and depth features.First,the left view saliency map is calculated by the improved graph-based manifold ranking algorithm.Second,the left view texture features and the depth features from stereoscopic images are extracted,and the texture and depth saliency maps are computed by the Log-Gabor filtering method,respectively.Third,the above three saliency maps are integrated into a stereoscopic(3D )saliency map by the weighted linear combination(WLC)method.Finally,the 3D saliency map is enhanced by the center-bias factor and visual acuity.The experimental results on a public eye tracking dataset show that the proposed model possesses a good detection performance and is superior to the existing 3D visual saliency detection models.