Optimal requantization of deep grayscale images and Lloyd-Max quantization

Area of Science:

• Computer Vision
• Digital Image Processing
• Signal Processing

Background:

• Classic signal quantization, introduced by Lloyd, is well-established for real-valued signals.
• Requantizing already digitized images presents unique challenges due to the discrete nature of the data.
• Existing quantization methods are often unsuitable for mapping high-bit-depth digital images to lower bit depths.

Purpose of the Study:

• To formulate and solve the problem of optimal mapping for digital grayscale images from a fine to a coarser bit scale.
• To evaluate the effectiveness of novel algorithms compared to traditional quantization methods in the digital domain.

Main Methods:

• Formulation of a novel digital quantization problem for high-bit-depth grayscale images.
• Experimental comparison of a dynamic programming algorithm against Lloyd's classic quantization method.
• Application to medical imaging scenarios (e.g., 9-13 bits-per-pixel to 8 bits-per-pixel).

Main Results:

• Classic quantization methods are found to be largely inapplicable to digital image requantization.
• A dynamic programming-based algorithm demonstrates significantly superior performance.
• The proposed method achieves better results in mapping digital images to coarser scales.

Conclusions:

• The discrete nature of digital images necessitates specialized requantization approaches.
• Dynamic programming offers a robust and effective solution for digital image scale mapping.
• This work provides a more suitable method for displaying high-resolution medical images on standard monitors.