Pixel-Domain Just Noticeable Difference Modeling with Heterogeneous Color Features.

Area of Science:

• Computer Vision
• Image Processing
• Human Visual Perception

Background:

• User-generated images necessitate advanced image compression techniques.
• Existing just noticeable difference (JND) models often neglect crucial color-oriented features impacting human visual system (HVS) perception.
• Understanding HVS characteristics like sensitivity, attention, and masking is vital for effective image compression.

Purpose of the Study:

• To develop a novel perception compression model for color images that fully incorporates HVS characteristics.
• To enhance existing JND models by integrating color-specific features and perceptual saliency.
• To improve noise concealment capacity in color image compression.

Main Methods:

• Extraction of local color-related features (color edge intensity, color complexity) and region-wise features (color area proportion, distribution position, dispersion).
• Introduction of 'color perception difference' as an inherent feature independent of color content.
• Modeling the interaction of features as 'color contrast intensity' and deriving 'color uncertainty' and 'color saliency'.
• Integration of color and uncertainty saliency into a JND model, accounting for masking and attention effects.

Main Results:

• The proposed model effectively integrates diverse color features and perceptual factors.
• Color uncertainty and saliency models are successfully applied to refine the conventional JND model.
• Subjective and objective experiments confirm the model's superior noise concealment capabilities.
• The enhanced JND model demonstrates improved performance over state-of-the-art methods.

Conclusions:

• The developed color perception model significantly advances image compression by incorporating detailed HVS characteristics.
• The model offers superior noise concealment, crucial for handling the influx of user-generated images.
• This research provides a more perceptually relevant approach to image compression, aligning with human visual processing.