EigenDenoise
Education
Only for Mac
Free
Mac
EigenDenoise is a native macOS application for image denoising and spectral analysis based on random matrix theory (RMT). It is designed for researchers, students, and practitioners working with high-dimensional image data, providing both classical and state-of-the-art denoising methods alongside interactive visualizations of the underlying mathematics.
KEY FEATURES
• Image Denoising — Two complementary algorithms run on the same noisy image stack:
– Classical Marchenko–Pastur (M-P) singular-value thresholding, with the standard spread rule from the batch MPPCA literature.
– Generalized-covariance oracle (rmt-denoise), modelling the bulk via the limiting spectral density of B_n = S_n T_n with a two-point spike weight H = β·δ_a + (1−β)·δ_1.
PSNR and rank are reported per method, side-by-side with clean / noisy / M-P / Gen-Cov reconstructions.
• Spectral Visualisation — Three interactive math tabs:
– Eigenvalue distribution: Monte-Carlo simulation of the spiked sample-covariance model, overlayed with the closed-form limiting density F_{y,H}(z) and the Case 1 / 2 / 3 support edges.
– Im(s) vs z: imaginary part of the Stieltjes transform across z.
– Roots vs β: cubic-root behaviour over the spike fraction.
• Curated Datasets — One-click downloads via the Datasets tab:
– ORL (AT&T) Faces — 400 grayscale 92×112 PGMs (40 subjects × 10 poses).
– CBSD68 — 68-image colour Berkeley test set (the classical denoise benchmark).
– Brain Tumour MRI — 3,264 T1 axial slices across four tumour classes (live GitHub listing).
Custom URL lists are supported, with per-image checkboxes and a destination-folder preview.
• Five Noise Models — Gaussian, Mixture-of-Gaussians, Two-Point H, Half-Gaussian, and Block-Half. Seedable for reproducibility.
• Metal GPU Acceleration — The Gram matrix multiply uses MPSMatrixMultiplication on Apple Silicon, with a transparent Accelerate (BLAS / LAPACK) fallback. Eigen-decomposition uses LAPACK dsyevd.
• Sandboxed Storage — Datasets save to your sandboxed Application Support folder by default; you may pick any local folder via security-scoped bookmark and the choice persists across launches.
EigenDenoise runs entirely on-device. There is no sign-in, no analytics, and no network communication except for the curated dataset downloads you initiate manually.
Ratings & Reviews
- This app hasn’t received enough ratings or reviews to display an overview.
Blind RIE denoising
A new optimal-shrinkage method that works without a clean reference image. It runs alongside the classical Marchenko-Pastur and generalized-covariance results, so you can compare all three on the same noisy stack and see the PSNR each one achieves.
Built-in "How to use" guide
Opens the first time you launch the app, and is always available from the ? button in the top right. It walks through the whole workflow -- getting images, choosing a test image and a method, reading the PSNR and rank figures -- and explains in plain language what the a, beta and y parameters actually mean.
Interface refinements across the Datasets and Image Processing tabs.
The developer, chienchung Yu, indicated that the app’s privacy practices may include handling of data as described below. For more information, see the developer’s privacy policy .
Data Not Collected
The developer does not collect any data from this app.
Accessibility
The developer has not yet indicated which accessibility features this app supports. Learn More
Information
- Seller
- chienchung Yu
- Size
- 3 MB
- Category
- Education
- Compatibility
Requires macOS 26.2 or later.
- Mac
Requires macOS 26.2 or later.
- Mac
- Languages
- English
- Age Rating
4+
- 4+
- Copyright
- © 2026 Yu Yao-Hsing
