This Python script compresses images by reducing the number of colors and optimizing the image representation. The idea of the program is to optimize scanned documents. It uses KMeans clustering to reduce the number of colors (higher contrast, smaller file size) and offers various options to customize the compression process. All supplied images are then saved as a multi-page PDF.
The color space of the input image (black-and-white / grayscale / color) is preserved as long as you use a local (per page) color palette. Otherwise, images with smaller color spaces will be adapted (black-and-white -> grayscale / color or grayscale -> color) to the color space of the image with the largest color space. For saving black-and-white images, I recommend having libtiff installed, otherwise it will fall back to using embedded jpeg.
This is a complete and improved rewrite of mzucker's noteshrink.
- Color Quantization: Reduces the number of colors in the document using KMeans clustering, leading to smaller file sizes and higher contrast.
- Background Detection and Removal: Identifies and removes the background color (replace with white), enhancing visual clarity.
- Autoremove empty pages: Identifies and removes blank pages.
- Customizable Palette: Allows you to specify the number of colors in the output palette, maximize saturation and choose between a global palette for all pages or individual palettes for each page.
- Denoising Options: Provides median filtering, morphological opening and unsharp masking to reduce noise and improve image quality.
- Parallel processing: Multi-threading for faster processing of multiple images.
- Low memory profile: Trys to keep memory usage as small as possible. This is done by:
- Each page is converted separately
- The intermediate files are stored in the current working directory (so that you can choose between disk and RAM [e.g.
/tmp]) - The library functions and programs used are specifically chosen to have the smallest overhead
- The input color space is preserved, if possible.
 |
 |
| Original Image | noteshrunk -w -s --unsharp_mask -ur 2 -ua 1 |
_NIPPON_YUSEN_KAISHA_-_S.S.KOBE_MARU_(at)_EN_ROUTE_(SS)_(NYPL_Hades-274833-469108).jpg){kind=link}
Morphological Opening swallows fine structures, but unsharp masking helps preserve them.
 |
Original Image (own) |
 |
noteshrunk -w --denoise_opening -os .9 |
 |
noteshrunk -w --unsharp_mask |
This page has a coverage of about 0.1 ‰ within the blue rectangle and will be removed, if --skip_empty is used.This is useful for duplex scanning. The margin width is 6% of the page width. |
 |
 |
 |
| Original Image | noteshrunk -w -s -tv 30 --denoise_opening -os 1.6 -n 6 -p 100 example_3-orig.jpg |
-
Compress a single image with default settings:
noteshrunk input.png
-
Compress multiple images with a white background and 16 colors:
noteshrunk -w -n 16 image1.jpg image2.png
-
Compress images using a local color palette and keep intermediate files while disabling multi-threading:
noteshrunk -l -j 1 -k *.jpg
- argcomplete
- NumPy
- Pillow (PIL Fork)
- Python 3
- scikit-image
- scikit-learn
- SciPy
- Ghostscript (executable
gsin PATH - for PDF merging)
libtifffor much smaller file sizes on black-and-white images.
pipx install noteshrunknoteshrunk [-h] [-o OUTPUT] [-w] [-s] [-n N_COLORS] [-d DPI] [-q [1-100]] [-l]
[-p PERCENTAGE] [-e] [-te THRESHOLD_EMPTY] [-j JOBS] [-y]
[-ts THRESHOLD_SATURATION] [-tv THRESHOLD_VALUE]
[--denoise_median] [--denoise_opening] [--unsharp_mask]
[-ms MEDIAN_STRENGTH] [-cs CLOSING_STRENGTH] [-ua UNSHARP_AMOUNT] [-ur UNSHARP_RADIUS]
[-k] [-v] [--version] files [files ...]
files: A list of paths to the input image files.-o,--output: Path to the output PDF file (default:output.pdf).-w,--white_background: Use white background instead of dominant color.-s,--saturate: Maximize saturation in the output image.-n,--n_colors: Number of colors in the palette (default: 8).-d,--dpi: DPI value of the input images (default: 300).-q,--quality: JPEG quality of the images embedded in the PDF (1-100, default: 75).-l,--local_palette: Create an individual color palette for each image (by sampling a -p percentage of the pixels of that image) instead of a global palette (by sampling a -p percentage of the pixels of each input image).-p,--percentage: Percentage of pixels to sample from each input image for color palette creation (default: 10).-e,--skip_empty: Pages with a coverage (after removing about 6 % at the margin) below-tewill be removed.-te,--threshold_empty: Coverage in parts per thousand / permille below which a page should be discarded.-j,--jobs: Number of threads to use for multi-threading (default: number of CPU cores).-y,--overwrite: Overwrite existing files without asking.-ts,--threshold_saturation: HSV saturation threshold (in percent) used for background detection (default: 15).-tv,--threshold_value: HSV value threshold (in percent) used for background detection (default: 25).--denoise_median: Apply median denoising on the output image with strength-ms.-ms,--median_strength: Strength of median filtering (default: 3).--denoise_opening: Apply morphological opening on the background mask with strength-os.-os,--opening_strength: Strength of opening filtering / radius of the structuring element (disk, default: 3).--unsharp_mask: Apply unsharp masking on the final image with radius-urand amount-ua.-ur,-ua: Radius and Amount used for unsharp masking.-k,--keep_intermediate: Keep the intermediate single-page PDFs.-v,--verbose: Verbose output.--version: Show program version and exit.
Contributions are welcome! Please feel free to submit issues or pull requests on the GitHub repository.
This project utilizes open-source software from the Python community. Special thanks to the developers and maintainers of the required libraries as well as mzucker's initial program.
This project is licensed under the MIT License. See the LICENSE file for details.