Vector Quantization (VQ) Image Coding Explorer

Compression Settings

Target Image Source

Vector Block Size (S × S)

Blocks of 2x2 pixels (dimension 4).

Target Bitrate per Vector 4 bits

Calculated on the fly...

Quantizer Clustering Method

What is Vector Quantization?

Vector Quantization (VQ) partitions an image into non-overlapping square blocks (vectors) of size $S \times S$. The encoder maps each block to the closest representative vector (code vector) in a codebook of size $K = 2^B$ ($B$ bits).

Only the index ($B$ bits) is transmitted, achieving a compression ratio of $(S^2 \times 8) / B$.

RMSE

Root Mean Square Error

PSNR

Peak Signal-to-Noise Ratio (dB)

Compression Ratio

Original vs Compressed

Original Image

256 × 256

Reconstructed Image (VQ Output)

Ready

Optimizing Codebook...

Bits Representation Analysis

Raw Image Size

524,288 bits

256 × 256 pixels @ 8 bits/pixel

Compressed Image Size

16,384 bits

Index size: 16,384 bits

Codebook Overhead Size

512 bits

16 vectors × 4 pixels × 8 bits

Code Vector Codebook Grid

Rendered at size 32×32 pixelated. Middle gray (128) is the zero-offset for mean-removed.

16 Vectors