基于改进四元傅里叶变换的显著性检测及其视频编码应用

Saliency detection based on improved quaternion Fourier transform and its application in video coding

针对基于感兴趣区域的有损视频压缩在低码率编码条件下容易产生明显的编码人工痕迹,提出一种基于注意力权重矩阵的四元傅里叶变换的视觉显著性视频编码模型。该方法引入人眼视觉注意力权重矩阵对不同区域图像四元数予以加权,该四元数由图像的亮度、色度和运动特征组成。图像视觉显著图可由其四元数特征的四元傅里叶相位谱获取。结合中心凹恰可觉察失真(FJND)模型将其应用于基于感兴趣区域视频编码,可提高视频编码质量。与五种流行的显著性检测算法在两个大型眼动跟踪数据库上进行对比实验,结果表明提出的算法显著性检测精度明显高于对比算法。此外,与最新的基于显著性视频编码方法在10段标准视频上进行编码视频的主观质量对比,该方法能提高低码率编码视频的主观视觉质量,且优于对比算法。

In order to reduce the undesirable compression artifacts for lossy video encoders with low bit rates,this paper proposed an improved saliency detection method based on phase spectrum of quaternion Fourier transform for ROI-based video coding. The method applied the human visual attention weight matrix to adjust the portion of the image quaternion characteristics consisting of intensity,colors and motion feature,and used the phase spectrum of the quaternion Fourier transform to obtain the saliency map of an image. Together with the foveated just-noticeable-distortion（ FJND） model,the new saliency detection algorithm was expected to achieve better visual quality for ROI-based video coding. Experimental results on two large benchmark databases demonstrate that the proposed method outperforms the five state-of-the-art methods in terms of accuracy.Moreover,by comparing with the newest saliency-aware video compression methods on 10 standard test video sequences,the proposed method can improve the performance in terms of subjective visual quality for low bit rates video coding.