Design and Analysis of a Video Compression System
Overview
This project focuses on the implementation and structural optimization of a custom video compression pipeline using Discrete Cosine Transform (DCT) spatial decorrelation, inter-frame motion estimation, and modern entropy coding architectures. Through rigorous experimentation across multiple pipeline variables, we analyze non-linear trade-offs between mathematical data fidelity (PSNR/MSE), computational throughput, and data compaction metrics (Compression Ratio).
Highlights
- Hybrid Video Architecture: Implementation of a standard temporal/spatial pipeline processing sequential intra frames (I-frames) via pure block-level DCT quantization alongside inter frames (P-frames) derived from motion-compensated residual matrices.
- Parametric Quantization Selection: Comparison of H.264/AVC standard matrix layouts against custom Human Visual System (HVS) weight structures optimized for empirical contrast sensitivity models.
- Adaptive Entropy Architectures: Performance profiling of variable-length Huffman codes against true range-restricted Arithmetic Coding, confirming an absolute efficiency increase of 1–2% approaching the absolute source entropy limit.
- Motion Estimation Optimization: Parametric evaluation of localized Sum of Absolute Differences (SAD) block-matching algorithms. Results demonstrate diminishing utility returns beyond an absolute spatial search range threshold of 8 pixels.
- Empirical Co-design Results: Identification of optimal global baseline variables featuring an $8\times8$ pixel pixel transformation block size, a baseline scaling multiplier of $\sigma = 1.0$, and range-restricted adaptive arithmetic serialization tables.
Code
- The code is available on GitHub