The fundamentals of average local variance--Part I: Detecting regular patterns

Area of Science:

• Image analysis and pattern recognition
• Remote sensing and geospatial analysis
• Computational imaging and computer vision

Background:

• The average local variance (ALV) method analyzes image spatial resolution.
• Previous hypotheses suggested ALV peaks relate directly to object size.
• The ALV method uses a moving window on coarsened images to generate a spatial resolution function.

Purpose of the Study:

• To explore the characteristics and peak locations of the ALV function.
• To test and refine the original hypothesis regarding ALV peak interpretation.
• To understand the relationship between ALV peaks and scene pattern structures.

Main Methods:

• Utilized synthetically generated image data for controlled analysis.
• Applied the average local variance (ALV) method with varying spatial resolutions.
• Investigated ALV function behavior across different synthetic scene structures.

Main Results:

• The original hypothesis that ALV peaks directly indicate object size is inadequate.
• ALV peak locations are primarily related to the geometric size of pattern structures, not always scene objects.
• Peaks can indicate object separation when distances exceed object size, and multiple peaks arise from complex structures.

Conclusions:

• A new hypothesis posits ALV peaks relate to geometric pattern structures, including object separation.
• ALV analysis is more complex than previously assumed, influenced by structure size, separation, and response mix.
• The ALV method can detect multiple structures but interpretation requires careful consideration of these complexities.