In the previous section Video Encoding Part-1, we looked at how individual images are represented and compressed using formats like BMP, PNG, and JPEG. However, video is not just a collection of independent images—it leverages compression both within each frame (spatial compression) and across frames (temporal compression) by exploiting similarities between consecutive frames.
In this article, we look at how images are actually represented in digital systems—from raw sensor data to common formats like RGB, YUV, JPEG, and PNG. Before we even talk about video, we need to understand how a single frame is stored, because video is just a sequence of these frames.
In Video Assessment Part 2, we looked at spatial quality metrics — how sharp, clear, and natural individual frames look. But for livestreaming or video conferencing, another dimension matters just as much: time.
(Even if each frame looks perfect, a video can still feel low-quality if motion is jerky, frames are dropped, or the playback stutters.)
In Video Quality Part 1 we introduced the basic idea of video quality metrics and the difference between reference-based and no-reference approaches.
In this part, we’ll go a bit deeper and look at how these methods work at a conceptual level, focusing on spatial quality—how individual video frames are evaluated for issues like blur, noise, and compression artifacts.
Video quality metrics like PSNR, SSIM, and VMAF were originally developed to evaluate video fidelity in controlled, offline settings, providing objective measurements of how closely a compressed or transmitted video matches its original.
In this series of articles on Video Quality Metrics, we’ll go through various approaches across different scenarios—ranging from offline comparison to live streaming, and from spatial to temporal quality assessment.