image processing

Bilingual

Perspec 1.0 – Desktop app for perspective correction of document photos
2 months ago
- Perspec 1.0 is a desktop app for correcting image perspective, especially for documents and receipts, now available on macOS, Windows, and Linux.
- The app uses a corner detection algorithm that segments photos into document and background, avoiding issues with curved or wrinkled edges common in other apps.
- It offers advanced binarization options, including 'Save BW Smooth' for anti-aliased black-and-white output, and handles shadows without complex local thresholding.
- Key updates in 1.0 include automatic corner detection, drag-and-drop support, improved selection UI, and EXIF rotation handling for PNGs.
- Perspec is built with Haskell and a custom C library (FlatCV) for image processing, emphasizing speed and accuracy over mobile app conveniences.
- The developer plans future features like fixed output sizes (A4/US Letter) and QR code detection for metadata automation.
G'mic 4.0: Squaring the Pixel, Easier
2 months ago
- G'MIC 4.0 marks a major milestone with 644 filters, focusing on stability and performance.
- New filters include Pixel Stretch, Gradients [Poles], Marker Drawing, Heightfield Warp, Offset Stripes, and Multi Color Transfer.
- The Multi Color Transfer filter is a novel algorithm from a research collaboration, supporting three methods and semi-supervised guidance.
- Core improvements include new commands like random_patches, shape_hilbert, pca, depthmap3d, and normalmap3d, plus enhanced matrix operations.
- A new repository, gmic-interactive-demos, offers interactive demonstrators for science outreach, such as the SteamFace Machine and Virtual Artist.
- G'MIC supports creative coding with short scripts and the 'G'MIC Adventures' series, showing rapid prototyping of visual effects.
- The project sees wider distribution through GEGL integration, PhotoFlare, Affinity Photo, and Snap Store availability.
Beej's Bit Bucket
2 months ago
- Image processing filters work using convolution matrices, where a small matrix (e.g., 3x3) is applied to each pixel and its neighbors to produce a new pixel.
- The process multiplies source pixel values by corresponding matrix elements, sums them, and divides by a divisor to compute the destination pixel.
- Different matrix values produce different effects: identity copies the image, lighten/darken adjust brightness, blur averages pixels, and sharpen uses weighted averages.
- Always write output to a separate destination buffer to avoid corrupting source pixels during computation.
- Edge handling can be done by cropping, assuming black/white, wrapping, or other methods; the demo masks edges with a border.
My Images Are Dithered
2 months ago
- Dithering reduces file size by using fewer colors while maintaining detail, and gives a cool aesthetic.
- The author uses ImageMagick's 'convert' command to simulate AM (amplitude-modulated) printing dots.
- The process involves converting to CMYK, separating channels, creating dot grids, rotating them at specific angles, and overlaying.
- Limiting each channel to 2 colors before combining produces a more accurate halftone effect.
- The old method limited the palette to monochrome (black and pink), while the new method uses a remap to a pink palette for more depth.
- The method is computationally expensive and not optimal for file size reduction, but achieves a distinctive print-like look.
- The script automates the process for multiple images, but the author notes it's not true dithering and can take 10 seconds per image.
Revisiting Yliluoma's ordered dither algorithm
3 months ago
- The article explores Joel Yliluoma's 2011 ordered dither algorithm and introduces simplified EMA variants.
- It compares these methods to Thomas Knoll's Photoshop dither algorithm, focusing on color selection for palette-based images.
- The EMA variants use exponential moving averages to pick candidate colors, with options for exact or constant mixing factors.
- Knoll's algorithm remains faster due to O(N log K) complexity, while Yliluoma-2 and EMA variants are O(NK).
- Dither quality depends on candidate count N and mixing factor t, with N around 2K recommended for K-color images.
- Source code is provided for implementations including Knoll, EMA, and offset dithers.
Show HN: Make images render brighter than white by abusing Rec.2100 PQ profiles
3 months ago
- SDR white (about 200 nits) is not a display's maximum brightness; HDR-capable panels can output 1,000–1,600 nits, leaving headroom.
- A Rec.2100 PQ color profile embedded in an 8-bit JPEG tricks color-managed renderers into displaying bright values in this headroom, making images glow brighter than the interface.
- Platforms like LinkedIn preserve ICC profiles through re-encoding, allowing the technique to work, though it fails with screenshots, re-saving, or compression.
- Superwhite properly processes images per pixel: decoding sRGB to linear light, converting primaries, masking white regions, boosting brightness, and encoding with PQ.
- The technique is a novelty with limited lifespan, as platforms may clamp brightness like they did for HDR video; subtle use is recommended for best effect.

image processing

Perspec 1.0 – Desktop app for perspective correction of document photos

G'mic 4.0: Squaring the Pixel, Easier

Beej's Bit Bucket

My Images Are Dithered

Revisiting Yliluoma's ordered dither algorithm

Show HN: Make images render brighter than white by abusing Rec.2100 PQ profiles