Frequency domain analysis and synthesis of image pyramid generating kernels

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Image pyramids are crucial for multi-resolution image analysis.
• Existing methods for generating image pyramids often lead to information loss.
• The optimal generating kernel for minimal information loss is related to an ideal low-pass filter.

Purpose of the Study:

• To compare frequently used generating kernels with an optimal kernel for image pyramid construction.
• To present physically realizable optimal generating kernels.
• To demonstrate methods for reducing computational cost in image pyramid generation.

Main Methods:

• Image pyramid construction as a two-dimensional decimation process.
• Comparison of common generating kernels against an ideal low-pass filter kernel.
• Utilizing half-band filters as components within the optimal kernel design.

Main Results:

• The optimal kernel ensures minimal information loss during resolution reduction.
• Physically realizable optimal kernels were successfully developed.
• Employing half-band filters significantly reduces the computational load.
• Image pyramids generated with the optimal kernel exhibit superior detail retention.

Conclusions:

• The proposed optimal kernel provides superior detail preservation in image pyramids.
• The use of half-band filters offers a computationally efficient approach.
• This research advances image pyramid generation for improved multi-resolution analysis.