What's wrong with mean-squared error?

By Bernd Girod. 1993

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Kede Ma
2. A. Bovik
3. Zhou Wang
4. Ke Gu
5. Weisi Lin
6. Lei Zhang
7. Dacheng Tao
8. Damon M. Chandler

Citations in the corpus

• Perceptual image quality assessment using a normalized Laplacian pyramid (2016, methodology): “Each pyramid coefficient is then divided by a local estimate of amplitude (weighted sum of absolute values of neighbors), where the weights are optimized for prediction of amplitude using (undistorted) images from a separate database.”
• Spatio-Temporal Just Noticeable Distortion Profile for Grey Scale Image/Video in DCT Domain (2009, background): “2013518 But unfortunately, they have been widely criticized for not correlating well with perceived quality measurement [1].”
• Color filter array demosaicking: new method and performance measures (2003, background): “It is generally agreed that current image measures cannot effectively quantify the perceptual quality of images obtained from a restoration process [14], [20], [21].”
• F Mean Squared Error: Love It or Leave It? (2009, background): “But is that the case? Unfortunately, the converse appears true when the MSE is used to predict human perception of image fidelity and quality [2]–[5].”
• 41 OBJECTIVE VIDEO QUALITY ASSESSMENT (2003, background): “However, they have been widely criticized as well for not correlating well with perceived quality measurement [1-8].”
• Image Quality Assessment for Fake Biometric Detection: Application to Iris, Fingerprint, and Face Recognition (2014, methodology): “somewhat controversial [45], [46], up to date, these are probably the most widely used methods for IQA as they conveniently make use of many known psychophysical features of the human visual system [47], they are easy to calculate and usually have very low computational complexity.”
• Just noticeable distortion model and its applications in video coding (2005, background): “The new model also accounts for the difference between edge regions and nonedge regions, since error is more visible in edge regions than in non-edge regions [9,11]; the formulae also applies to color components.”
• Perceptual Video Coding Based on SSIM-Inspired Divisive Normalization (2013, background): “However, existing video coding techniques typically use the sum of absolute difference (SAD) or sum of square difference (SSD) as the model for distortion, which have been widely criticized in the literature for the lack of correspondence with perceptual quality [1]–[3].”
• Perceptual visual quality metrics: A survey (2011, background): “It has been well acknowledged that a signal fidelity measure does not align well with human visual perception of natural images or computer generated graphics [41,52,97,149,161].”
• Why is image quality assessment so difficult? (2002, background): “In the past three to four decades, a great deal of effort has been made to develop new objective image and video quality measurement approaches which incorporate perceptual quality measures by considering human visual system (HVS) characteristics [1, 2, 3, 4, 5, 6, 7, 8, 9].”
• Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment (2016, background): “Although being widely used, it does not correlate well with perceived visual quality [6].”
• Image quality assessment: from error visibility to structural similarity (2004, unknown): “But they are not very well matched to perceived visual quality (e.g., [1]–[9]).”