Peak Signal-to-Noise Ratio (PSNR)

INFO

Measures the quality of reconstructed images compared to the original, used in image processing.

How It Works

PSNR is derived from the Mean Squared Error (MSE) between the original image $I$ and the reconstructed image $K$:

$$\text{PSNR}=10\cdot {\log }_{10}\left(\frac{MA{X}_I^2}{\text{MSE}}\right)$$

• $MA{X}_I$: Maximum possible pixel value (e.g., 255 for 8-bit images)
• $\text{MSE}$: Mean squared error between $I$ and $K$

Higher PSNR indicates better reconstruction fidelity.

What to Look For

• Typical range: 30–50 dB for 8-bit images
• Higher PSNR = lower distortion
• Sensitive to pixel-level differences, but not perceptual quality
• Best used when comparing similar codecs or reconstruction methods

Application Models

• Autoencoders
• Generative Adversarial Networks (GANs)
• Convolutional Neural Network (CNN)