Abstract
Recently introduced inpainting algorithms using a combination of applied harmonic analysis and compressed sensing have turned out to be very successful. One key ingredient is a carefully chosen representation system which provides (optimally) sparse approximations of the original image. Due to the common assumption that images are typically governed by anisotropic features, directional representation systems have often been utilized. One prominent example of this class are shearlets, which have the additional benefit of allowing faithful implementations. Numerical results show that shearlets significantly outperform wavelets in inpainting tasks. One of those software packages, ShearLab, even offers the flexibility of using a different parameter for each scale, which is not yet covered by shearlet theory. In this paper, we first introduce universal shearlet systems which are associated with an arbitrary scaling sequence, thereby modeling the previously mentioned flexibility. In addition, this novel construction allows for a smooth transition between wavelets and shearlets and therefore enables us to analyze them in a uniform fashion. For a large class of such scaling sequences, we first prove that the associated universal shearlet systems form band-limited Parseval frames for $L^2(\mathbb{R}^2)$ consisting of Schwartz functions. Second, we analyze the inpainting performance of this class of universal shearlet systems within a distributional model situation using an $\ell^{1}$-analysis minimization algorithm for reconstruction. Our main result states that, provided that the scaling sequence is comparable to the size of the (scale-dependent) gap, asymptotically perfect inpainting is achieved.
Introduction
Image inpainting—the process of recovering missing or corrupted parts of an image—is a cornerstone of digital restoration and medical imaging. Traditionally, researchers have used wavelets to represent images. While wavelets are excellent at capturing isotropic features ("blobs"), they struggle with anisotropic features like edges because they lack directionality.
Shearlets were developed to address this by using elongated, directional atoms. However, standard shearlet theory often assumes a fixed "parabolic scaling." This research addresses two main gaps: the lack of theoretical flexibility to change anisotropy across scales and the need for a rigorous mathematical understanding of why shearlets outperform wavelets in $\ell^1$-minimization inpainting tasks.
Universal Shearlet Systems
The authors introduce Universal Shearlet Systems, a unified framework that replaces a fixed scaling parameter with a scaling sequence $(\alpha_j)$. This allows the system to behave like wavelets at some scales and like highly directional shearlets at others, providing a smooth transition between different representation types.
The paper proves that for a large class of scaling sequences, these systems form Parseval frames, providing a stable and redundant way to decompose signals. Furthermore, these systems consist of band-limited Schwartz functions, ensuring excellent localization in both space and frequency.
Theoretical Analysis of Inpainting
To evaluate inpainting performance, the authors use an $\ell^1$-analysis minimization algorithm. Instead of building a signal from atoms (synthesis), they seek a reconstructed signal whose shearlet coefficients are as sparse as possible while remaining consistent with the observed data.
The performance is analyzed using a distributional model: a sharp edge (represented by a line distribution) interrupted by a vertical gap of width $2h_j$. This model is motivated by real-world scenarios such as seismic data acquisition, where sensor failures often result in missing strips of data.
Key Results: Asymptotic Perfection
The main theoretical contribution is the proof of asymptotically perfect inpainting. The result states that if the gap width $h_j$ shrinks faster than the length of the shearlet elements at fine scales, the recovery error converges to zero.
This formally confirms that the success of inpainting depends on the relationship between the gap size and the anisotropy of the representation. Because shearlets are long and thin, they can "reach across" gaps that would be insurmountable for isotropic wavelets. By using "clusters" of shearlets oriented along the edge, the system can effectively ignore the gap and focus on structural continuity.
Conclusion
This work provides the first theoretical backbone for the observed superiority of shearlets over wavelets in inpainting. By bridging the gap between theory and practice, it shows that highly anisotropic systems are essential for recovering curvilinear structures. While the current proof focuses on straight lines and noiseless models, it sets the stage for future extensions to curved edges and noisy data.
Summary updated