Quantization table design revisited for image/video coding.

Area of Science:

• Digital image and video compression
• Information theory and coding
• Signal processing

Background:

• Traditional quantization table design is often coupled with specific encoding methods.
• Soft decision quantization (SDQ) offers potential for improved compression efficiency.
• Optimizing quantization tables independently for reconstruction quality is a key challenge.

Purpose of the Study:

• To develop a novel quantization table design methodology for image/video coding, specifically considering soft decision quantization (SDQ).
• To optimize quantization tables for reconstruction purposes, assuming optimal SDQ encoding.
• To propose an efficient algorithm for designing these tables based on statistical models.

Main Methods:

• Modeling transform coefficients as independent random sources and applying the Shannon lower bound to approximate rate-distortion functions.
• Developing a statistical-model-based algorithm using the Laplacian model for quantization table design in DCT-based image coding.
• Evaluating the proposed algorithm's performance against standard JPEG encoding and state-of-the-art optimizers.

Main Results:

• Achieved over 1.5-dB PSNR gain in standard JPEG encoding with negligible complexity increase.
• Outperformed a state-of-the-art JPEG optimizer by 0.5-dB PSNR with over 2000x reduced complexity (SDQ OFF).
• Provided 0.2-dB+ PSNR gain with 85% complexity reduction (SDQ ON), demonstrating broad applicability.

Conclusions:

• The proposed quantization table design principle, assuming optimal SDQ, enables effective optimization for reconstruction quality.
• The statistical-model-based algorithm offers significant compression performance improvements across various image coding systems.
• This approach provides a more efficient and effective method for quantization table design in image and video compression.