PSNR (Peak signal-to-noise ratio) is the simplest and most common metric for assessing objective image quality. PSNR is a ratio of the maximum possible signal to the noise level (in dB). Noise is defined via mean squared error (MSE) between two images - original image and its noisy approximation. For 8 bit images, the maximum possible pixel value is 255. Typical PSNR values for image compression are in the range of 30 to 50 dB. The higher the metric, the better the image quality. For video, the formula differs only in that the noise level is calculated as the sum of the noise level between the corresponding frames of the original video and the compressed one.
PSNR-HVS. Good PSNR does not always guarantee good quality, due to the fact that the human visual system (HVS) has non-linear behavior. Two images can have the same PSNR and completely different subjective quality scores. Especially images corrupted with additive Gaussian noise, high-frequency noise, impulse noise, Gaussian blur etc. HVS is more sensitive to low-frequency distortions rather than to high-frequency ones. It is also very sensitive to contrast changes and noise. For PSNR-HVS noise level is defined via mean square errors weighted in the DCT domain with specific weights. Any contrast change and mean shift lead to non-zero MSE and, as a result, small PSNR-HVS.
Aligned-PSNR (APSNR) is applied for an objective assessment of the quality of video stream over wireless and mobile network. Due to packet loss problems in wireless and mobile networks using conventional PSNR leads to distorted results because when calculating the noise level, not matching frames are compared. APSNR allows you to correctly match frames if there is packet loss. APSNR algorithm is divided into three parts - corresponding frames searching, shifting the streamed video frame, and calculating PSNR value among the pairs of the frames. Searching. For the current frame of streaming video, the PSNR is computed with each frame of the original video inside the fixed window. A streaming video frame and an original video frame with maximum PSNR are considered as corresponding frames. Shifting. If the maximum PSNR is not obtained for the first pair of frames, then there is packet loss and you need to move the streaming video forward by the corresponding number of frames. Then go to the next frame of the streaming video and repeat steps 1 and 2 until all frames are passed. Finally, calculate the average PSNR for all found frame pairs.