Adaptive downsampling to improve image compression at low bit rates.

Area of Science:

• Digital image processing
• Image compression algorithms
• Computer vision

Background:

• Traditional image compression methods like JPEG can struggle to maintain quality at very low bit rates.
• Downsampling an image before compression and reconstructing it later can improve coding efficiency.

Purpose of the Study:

• To develop a novel image compression algorithm that enhances coding quality at low bit rates.
• To improve upon existing downsample-prior-to-compression schemes by adaptively selecting parameters.

Main Methods:

• The algorithm adaptively selects downsampling directions, ratios, and quantization steps based on local visual significance.
• It restores full-resolution images directly from DCT coefficients of downsampled pixels, avoiding spatial interpolation.
• The method integrates adaptive mode decision for seamless switching to standard JPEG at higher bit rates.

Main Results:

• The proposed algorithm significantly increases the critical bit rate to approximately 1.2 bpp, compared to 0.15-0.41 bpp for existing methods.
• It demonstrates superior Peak Signal-to-Noise Ratio (PSNR) improvement over existing techniques below the critical bit rate.
• The adaptive approach reduces image dependency of the critical bit rate and automates coder switching.

Conclusions:

• The novel adaptive algorithm offers superior image coding quality at low bit rates.
• It outperforms standard JPEG over a much wider bit-rate scope.
• The adaptive nature makes it suitable for variable bit-rate applications.