The structure of two-dimensional scalar fields with applications to vision

Area of Science:

• Computer Vision
• Image Analysis
• Mathematical Imaging

Background:

• Traditional image analysis relies on linear decomposition of 2D scalar fields, which often proves unnatural and lacks invariance to topological deformations.
• Existing methods fail to preserve perceptual structure and can result in decomposed parts more complex than the original image.

Purpose of the Study:

• To develop a novel method for decomposing 2D scalar fields that respects their inherent structure.
• To introduce a hierarchical decomposition that generates featureless components, invariant to topological and grayscale deformations.

Main Methods:

• A hierarchical decomposition method is presented, representing images as a superposition of featureless parts (lacking local extrema or saddle points).
• The hierarchical structure is conceptualized as a generative grammar for smooth pictures.
• The method is extended to analyze pictures sampled with graded apertures, introducing the concept of an aperture spectrum.

Main Results:

• The proposed decomposition yields featureless components, overcoming the limitations of unnatural linear superpositions.
• The hierarchical structure provides a generative grammar for smooth images.
• The aperture spectrum offers a new way to describe image structure based on sampling apertures.

Conclusions:

• The novel hierarchical decomposition offers a more natural and perceptually relevant way to analyze 2D scalar fields compared to traditional linear methods.
• This approach generates a robust representation invariant to common image transformations.
• The aperture spectrum is a promising tool for analyzing visual functions and image structures sampled under varying conditions.