Nonlinear and Nonseparable Bidimensional Multiscale Representation Based on Cell-Average Representation

Area of Science:

• Digital image processing
• Multiscale analysis
• Nonlinear signal processing

Background:

• Traditional multiscale representations often struggle with piecewise continuous functions.
• Existing methods may not efficiently adapt to local function characteristics.
• Image compression and super-resolution require advanced representation techniques.

Purpose of the Study:

• To develop a new nonlinear and nonseparable multiscale representation for bidimensional functions.
• To introduce adaptivity into the representation through a nonlinear prediction operator.
• To demonstrate the utility of this representation in image processing applications.

Main Methods:

• Construction of a nonlinear and nonseparable multiscale representation.
• Definition of a linear projection operator.
• Incorporation of a locally adaptive nonlinear prediction operator.
• Application to image compression and super-resolution.

Main Results:

• The proposed representation effectively handles piecewise continuous bidimensional functions.
• The adaptive prediction operator leads to a significant reduction in significant coefficients.
• Demonstrated improvements in image compression efficiency.
• Successful application to enhance image super-resolution.

Conclusions:

• The developed nonlinear multiscale representation offers advantages over existing methods.
• Its adaptivity is key to achieving efficient image compression.
• The representation shows promise for advanced image restoration tasks like super-resolution.