MSSSIM

MSSSIM, or Multi-Scale Structural Similarity Index, is an objective image quality metric that extends the Structural Similarity Index (SSIM) to multiple resolution scales. It measures how similarly two images or video frames appear, taking into account perceptual factors such as luminance, contrast, and structural information. By operating across scales, MSSSIM aims to capture image quality degradation that can be introduced by compression, filtering, or transmission, which may manifest differently at different resolutions.

Algorithmically, the method builds a Gaussian pyramid to produce a set of progressively downsampled versions of the images. For each scale, local measurements of luminance are computed, and for the intermediate scales, contrast and structure are evaluated. The final MSSSIM score is formed by aggregating the scale-wise comparisons, typically as a weighted product of luminance, contrast, and structure terms across all scales. The score lies in [0,1], with 1 indicating perfect similarity.

Usage and considerations: MSSSIM is widely used to assess perceived image quality in image and video processing applications, such as compression, denoising, and streaming. Compared with SSIM, MSSSIM can better account for artifacts that appear at different resolutions. It is commonly applied to grayscale images but can be extended to color by applying the metric to luminance and chrominance channels separately or in a joint manner.

Limitations: While MSSSIM aligns well with human perception in many cases, it is not a perfect proxy for subjective quality across all content. It requires a reference image for full assessment and can be computationally more intensive than single-scale metrics.