A lightweight Python implementation of IW-SSIM that balances accuracy and speed.

This metric enhances the traditional MS-SSIM by using an information-theoretic weighting strategy, prioritizing areas of the image that contain more visual information where the human eye is more sensitive to distortions.

Traditional SSIM treats all pixels equally. IW-SSIM, however, applies Information Content (IC) weighting. It calculates a "weight map" based on a Gaussian Scale Mixture (GSM) model, ensuring that complex structural regions have a higher impact on the final score than flat, uninformative areas.

• Laplacian Pyramids for multi-scale analysis.
• Structure Tensors to evaluate local image statistics.
• Perceptual Color Space: The code linearizes RGB values and converts them to the perceptual luminance ($L^*$) of the CIELAB space for better alignment with human vision.

While based on the original research by Wang and Li, simple-iw-ssim introduces several practical modifications:

• Simplified Information Map: Instead of a full GSM-based statistical model, this version uses a streamlined structure tensor approach to calculate the information distribution.
• Linearized Workflow: Unlike many basic implementations that work directly on gamma-corrected sRGB, this tool performs linearization and uses perceptual luminance ($L^*$).
• Coherence-Adjusted Weighting: Eigenvalues are scaled by their coherence to slightly boost the relative importance of structured edges compared to chaotic textures.

Built only on NumPy, SciPy, and Pillow. No need for massive frameworks like OpenCV or PyTorch.

python iwssim.py reference.png distorted.png