TL;DR
This paper introduces Disk Haar Functions (DHFs), a novel set of orthogonal functions specifically designed for analyzing circular images. By overcoming the "edge effect" inherent in traditional methods, DHFs offer a more accurate and efficient way to represent and process images with circular regions, such as interference patterns and lunar craters.
What Problem Does This Paper Address?
Analyzing circular images, like interference patterns or lunar craters, presents a challenge for standard image processing techniques. Traditional methods often rely on 2D tensor product orthogonal functions (TPOFs) that operate on rectangular grids. When applied to circular images, these methods require extending the image to a rectangular region, leading to an "edge effect" that degrades reconstruction quality. This paper aims to address this limitation by developing a new class of functions tailored for circular domains.
Methods
The core contribution is the introduction of Disk Haar Functions (DHFs). These functions are derived from traditional Haar functions but are defined on a unit disk. The process involves:
- Equal-area Partitioning: The unit disk is divided into a series of concentric rings and then each ring is further divided into sectors. This partitioning ensures that each sub-region has an equal area.
- Mapping to TPHFs: A one-to-one mapping is established between these disk sub-regions and the sub-regions of a 2D tensor product Haar function (TPHF) grid. This mapping allows the construction of DHFs.
- Preprocessing for Digital Images: Since digital images are typically represented on a Cartesian grid, a preprocessing step involving image interpolation (specifically bicubic interpolation) is used to convert Cartesian pixel data into a polar representation suitable for DHFs.
- DHF Transform: The paper outlines the decomposition and reconstruction algorithms using DHFs, analogous to standard transforms.
Datasets
The paper uses two types of datasets for evaluation:
- Optical Interference Image: A specific interference pattern image is used to demonstrate the reconstruction capabilities of DHFs compared to FFT.
- Lunar Crater Images: Ten CCD images of lunar craters captured by Chang'e-1 are used to analyze the effectiveness of DHFs for feature extraction (energy calculation) and classification.
Key Insights and Ideas
- The "Edge Effect" is a Real Problem: Traditional methods struggle with circular regions due to the artificial extension to rectangular grids, leading to noticeable artifacts in reconstructions.
- DHFs are Tailored for Circularity: By defining functions directly on a disk and using an equal-area partitioning, DHFs naturally handle the geometry of circular images.
- Improved Reconstruction Quality: DHFs significantly outperform FFT in reconstructing circular images, especially when using a comparable number of coefficients.
- Feature Extraction for Craters: The energy of a crater, calculated using DHFs, shows a significant increase for a certain range of coefficients and then stabilizes, suggesting it can be a useful feature for crater classification and recognition.
Key Results
- Interference Pattern Reconstruction: Reconstructed images using DHFs show much better quality than those reconstructed using FFT, with fewer artifacts, even when using the same number of decomposition coefficients.
- Lunar Crater Analysis: The energy of lunar craters, when represented by DHFs, exhibits a clear trend of increasing and then stabilizing with an increasing number of coefficients. This energy can differentiate between craters, offering a potential for classification. The paper implies that DHFs provide a more robust representation for this analysis.
Limitations and Future Work
- Preprocessing Dependency: The effectiveness of DHFs relies on an accurate image interpolation procedure to convert Cartesian images to a polar representation. The paper mentions bicubic interpolation but doesn't delve deeply into its limitations or alternatives.
- Future Work: The authors suggest further investigation into the applications of DHFs in lunar crater analysis, particularly exploring the relationship between frequency spectrum and geological attributes like crater age. They also aim to study the connection between frequency spectrum and geological attributes.
Summary updated