基于视觉掩盖效应和奇异值分解的图像质量评测方法  

Image quality evaluation based on visual masking effect and singular value decomposition

作　　者：袁飞[1] 黄联芬[1] 姚彦[1] 

机构地区：[1]厦门大学信息科学与技术学院通信工程系,福建厦门361005

出　　处：《光学精密工程》2008年第4期706-713,共8页Optics and Precision Engineering

基　　金：广州京信研究基金资助项目(No.K81004)

摘　　要：将人眼视觉特性和图像本征特性相结合,设计了一套基于视觉掩盖和奇异值分解的图像质量评测方法。根据人眼的视觉敏感性函数(CSF)将参考图像和待测图像做噪声掩盖处理,剔除视觉敏感区外的图像信息;然后,根据人眼视觉特性提取参考图像的纹理区和平坦区,并作为待测图像检测区的划分依据,分别计算参考图像和待测图像在各自纹理区和平坦区的奇异值;最后,通过奇异值测度得到待测图像在纹理区和平坦区的质量劣化程度。实验表明:通过纹理区和平坦区的区分检测,可提高检测敏感度并据此判断劣化及影响程度。同时,检测结果的主观相似性(CC)可达到0.896,均方根误差(RMSE)为5.58,较PSNR和SSIM具有更好的性能。该方法实现简单,便于系统集成,可在涉及视频及图像质量检测的领域得到较好应用。A image quality evaluation method based on visual masking effect and singular value decomposition is proposed. According to the Contrast SensibilityFunction （CSF）, the original image and the processed image are preprocessed to remove the signals outside the sensible region. On the basis of the property of the Human Visual System （HVS）, the texture and plat regions of the original image are extracted to used as a division standard. Then, the singular values of the original and processed images are calculated both in texture and plat regions. Finally, by measuring the difference between the singular value metrics of the original and the processed images, the image quality can be evaluated. The experimental results show that this method is sensible to the errors perceived by human-being and can improve discrimination for the distortion region. The correlation between the result and the subjective score is 0. 896 and the root mean square error （RMSE） is 5.58,which shows the method can be realized easily and can be embedded into the application system.

关 键 词：图像质量 奇异值分解 视觉掩盖 客观评价 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]