论文部分内容阅读
为进一步提高现有视频编码技术的压缩效率及解码重建图像的主观视觉感知质量,在现有人眼恰可察觉失真(JND,just noticeable distortion)模型的基础上,提出了恰可察觉编码失真(JNCD,just noticeable coding distortion)模型。首先,通过主观实验,对恰可察觉梯度幅值差异(JNGD,just noticeable gradient difference)进行了研究,分析其变化规律并建立JNGD模型。使用全变分(TV,total variation)方法将图像分解为结构图和纹理图后,分别求取其梯度信息得到结构梯度图和纹理梯度图,利用JNGD模型分别滤除结构梯度图和纹理梯度图中的人眼不可察觉的梯度幅值;其后,分析了人眼感知对于不同梯度幅值的编码失真敏感性,设计了梯度幅值与JNCD值的主观实验,得到两者的关系模型;最后,考虑人眼对图像中的边缘、平坦和纹理3类区域失真感知程度的差异性,利用滤波后的结构梯度和纹理梯度信息将图像划分为上述3类区域,最终建立整幅图像的JNCD模型。为验证本文提出的JNCD模型的可靠性,在高效视频编码(HEVC)标准测试平台上进行的模型验证结果表明,在本模型指导下的编码其解码重建图像获得了较好的主观视觉效果,可为人眼视觉感知冗余的分析及感知编码的改进提供依据。
In order to further improve the compression efficiency of existing video coding and the subjective visual perception quality of decoded reconstructed images, based on the existing JND (just noticeable distortion) model, a novel method called just perceptible coding distortion (JNCD , just noticeable coding distortion model. First of all, through subjective experiments, we study the just noticeable gradient difference (JNGD), analyze its variation and establish JNGD model. Using the TV (total variation) method to decompose the image into a structure map and a texture map, the gradient information is obtained separately to obtain the structure gradient map and the texture gradient map, and the structure gradient map and the texture gradient map are respectively filtered by using the JNGD model Then, we analyzed the sensitivity of the human eye to the coding distortion of different gradient magnitudes, designed the subjective experiment of gradient magnitude and JNCD value, and obtained the relational model between the two. Finally, , Taking into account the human eye’s perception of distortion in the three types of regions of the edge, flatness and texture in the image, the image is divided into the above three types of regions using the filtered structural gradient and texture gradient information, and finally the JNCD model of the entire image is established . In order to verify the reliability of the JNCD model proposed in this paper, the model verification results on the standard test platform of High Efficiency Video Coding (HEVC) show that the subjective visual effect obtained by decoding the reconstructed image under the guidance of this model can obtain good subjective visual effects Provide the basis for the analysis of human visual perception redundancy and the improvement of perceptual coding.