[Paper Review] Image Separation using Wavelets and Shearlets

Gitta Kutyniok, Wang‐Q Lim|arXiv (Cornell University)|Jan 3, 2011

Medical Image Segmentation Techniques14 references4 citations

TL;DR

This paper proposes a novel image separation method that uses a combined dictionary of translation-invariant wavelets and compactly supported shearlets to sparsely represent point- and curvilinear features, respectively, leveraging ℓ¹ minimization for geometric separation. The approach achieves superior accuracy and speed over MCALab—especially for highly curved structures—due to shearlets' excellent spatial localization, validated both visually and quantitatively with real neurobiological images.

ABSTRACT

In this paper, we present an image separation method for separating images into point- and curvelike parts by employing a combined dictionary consisting of wavelets and compactly supported shearlets utilizing the fact that they sparsely represent point and curvilinear singularities, respectively. Our methodology is based on the very recently introduced mathematical theory of geometric separation, which shows that highly precise separation of the morphologically distinct features of points and curves can be achieved by $\ell^1$ minimization. Finally, we present some experimental results showing the effectiveness of our algorithm, in particular, the ability to accurately separate points from curves even if the curvature is relatively large due to the excellent localization property of compactly supported shearlets.

Motivation & Objective

• To address the challenge of separating point- and curvilinear features in images, a task critical in neurobiology and astronomy.
• To overcome limitations of existing methods like MCALab, which rely on curvelets and struggle with high-curvature structures.
• To leverage the superior directional and spatial localization properties of compactly supported shearlets for improved morphological separation.
• To develop a faster, more accurate algorithm based on a mathematically grounded framework of geometric separation.
• To provide a freely available, reproducible implementation within the ShearLab toolbox for broader scientific use.

Proposed method

• The method employs a combined dictionary of translation-invariant wavelets and compactly supported shearlets to sparsely represent point- and curvilinear singularities.
• It formulates the image separation as an ℓ¹ minimization problem over the expansion coefficients of the image in the combined wavelet-shearlet dictionary.
• The algorithm uses block relaxation to efficiently solve the optimization problem, alternating between updating the pointlike and curvilinear components.
• Shearlets are used due to their optimal sparse representation of cartoon-like images and superior spatial localization compared to band-limited curvelets.
• The approach is grounded in the mathematical theory of geometric separation, which guarantees high-precision decomposition under suitable conditions.
• The implementation is integrated into the ShearLab toolbox, ensuring reproducibility and accessibility for the research community.

Experimental results

Research questions

• RQ1 Can a combined dictionary of wavelets and shearlets achieve more accurate separation of point- and curvilinear features than existing methods based on wavelets and curvelets?
• RQ2 Does the superior spatial localization of compactly supported shearlets lead to improved separation performance, especially for highly curved structures?
• RQ3 Can ℓ¹ minimization over a wavelet-shearlet dictionary achieve geometric separation with high precision, as theoretically predicted?
• RQ4 Is the proposed algorithm faster and more robust than MCALab, particularly in handling complex curvilinear features?
• RQ5 Can the method be effectively applied to real-world biomedical images, such as neuron fluorescence microscopy data?

Key findings

• The proposed algorithm outperforms MCALab in separating point- and curvilinear features, particularly for structures with high curvature, due to the enhanced spatial localization of compactly supported shearlets.
• Quantitative evaluation shows that the proposed method achieves lower error in both pointlike and curvilinear component reconstruction across all tested thresholds (0 < T < 1), as measured by Mₚ and M_c metrics.
• The algorithm runs in 135.37 seconds on average (15 iterations), significantly faster than MCALab’s 182.19 seconds (30 iterations), demonstrating improved computational efficiency.
• Visual comparisons confirm that the proposed method better preserves curved structures, with fewer artifacts and less leakage of curve parts into the point component.
• In neurobiological applications, the method successfully separates spines (pointlike) and dendrites (curvilinear) in real fluorescence microscopy images, validating its practical utility.
• The ShearLab implementation of the algorithm is publicly available, supporting reproducibility and broader adoption in the research community.