A perceptually relevant MSE-based image quality metric.

Area of Science:

• Image Processing
• Computer Vision
• Perceptual Quality Assessment

Background:

• Traditional image quality metrics (IQMs) like Mean Squared Error (MSE) and Structural Similarity Index (SSIM) are widely used but may not fully capture perceived visual quality.
• Analyzing the relationship between MSE and SSIM under additive noise provides insights for developing more perceptually relevant metrics.

Purpose of the Study:

• To propose a novel MSE-based image quality metric, MSE-SSIM, that is perceptually relevant and computationally efficient.
• To evaluate the performance of MSE-SSIM against existing IQMs using public image databases.
• To demonstrate the applicability of MSE-SSIM in image and video optimization tasks, specifically in Wiener filter design.

Main Methods:

• Developed MSE-SSIM by analyzing the relationship between MSE and SSIM under an additive noise distortion model.
• Expressed MSE-SSIM in terms of source image variance and the MSE between source and distorted images.
• Evaluated MSE-SSIM performance on LIVE, CSIQ, and TID2008 image databases.

Main Results:

• MSE-SSIM demonstrates favorable performance compared to several existing IQMs, despite requiring less computation.
• The proposed metric shows improved performance in perceptual quality assessment tasks.
• Images filtered using a MSE-SSIM-optimal Wiener filter exhibit superior visual quality compared to those filtered with a MSE-optimal Wiener filter.

Conclusions:

• MSE-SSIM is a computationally efficient and perceptually relevant image quality metric.
• Its simplicity makes it suitable for various image and video optimization problems.
• The metric effectively enhances visual quality in applications like Wiener filtering.