IPAD: Intensity Potential for Adaptive De-Quantization.

Area of Science:

• Computer Vision
• Image Processing
• Display Technology

Background:

• High bit-depth displays (≥10 bits) are common, but mainstream media uses lower bit depths (8 bits).
• De-quantization is needed to bridge this gap, but traditional methods like zero-padding cause false contour artifacts.
• Existing filtering and interpolation methods use limited neighbor information, failing to eliminate these artifacts effectively.

Purpose of the Study:

• To propose a novel intensity potential (IP) field for modeling complex pixel relationships in images.
• To develop an adaptive de-quantization procedure based on the IP field for converting low bit-depth images to high bit-depth.
• To address the limitations of traditional de-quantization methods and improve image fidelity.

Main Methods:

• A novel intensity potential (IP) field is introduced, where potential decreases with spatial distance and potentials are additive.
• An adaptive de-quantization algorithm is developed using the proposed IP field to estimate least-significant bits.
• The method was tested on natural, synthetic, and high-dynamic range image datasets.

Main Results:

• The proposed IP field effectively models complex pixel interdependencies and preserves local consistency.
• The adaptive de-quantization procedure significantly reduces false contour artifacts.
• Experiments show substantial improvements in peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) over state-of-the-art methods.

Conclusions:

• The novel intensity potential field offers an effective approach for image de-quantization.
• This method provides a significant advancement in converting low bit-depth images for high bit-depth displays.
• The IP field approach demonstrates superior performance and artifact reduction compared to existing techniques.