* Add rainbow_artifacts_eraser helper
This helper file contains the methods necessary to perform a fourier transform on the picture, to remove frequencies responsible for rainbow artifacts on Kaleido screens, and performe the reverse fourier transform
* Replace blurring method with frequency removal method to erase rainbow effect on Kaleido 3 screens
* High performance improvements by using rfft2 instead of fft2
* Fine-tuned the settings and added the perpendicular direction for a better final rendering
The finer settings allow for more information to be retained in the final image, while still effectively removing the rainbow effect.
Adding the perpendicular direction results in a better rendering of the final image (avoiding visual artifacts related to suppression at the main angle).
* Revert the addition of perpendicular angles and lower attenuation_factor
It was a mistake to add the perpendicular angles in the previous commit: I had the rainbow effect removal process called 2 times when I did this, for testing purposes (One before downscale and one after downscale).
The proper way to call the process is only after the downscale. And in this case it is not necessary to remove frequencies along the perpendicular angles.
In the mean time, attenuation_factor=0.15 has proven to work well along a collection of testing images.
It should be my latest commit for this feature
* Also attenuate high frequencies at 45°
CFA is sometimes orientated at 135°, sometimes at 45° so until we find if there is a law depending on the screen size, e-reader model or something, the best we can do is attenuate high frequencies on those two directions
* fix imports
* Update comic2ebook.py
Calculate is_color with (opt.forcecolor and img.color)
pass is_color to img.optimizeForDisplay
* Update image.py
Remove color check condition, because now we process colored images too.
Pass is_color to erase_rainbow_artifacts
* Update rainbow_artifacts_eraser.py
Add support for colored images: Convert to YUV, extract luminance channel, do FFT -> Filter -> IFFT on luminance channel, insert back to YUV, convert back to RGB
To maximize compatibility until we know for sure the orientation of CFA for each device, filtering is now done on 135° + 45° axis
After more testing, attenuation_factor is decreased to 0.10
* Update comic2ebook.py
Rename rainbow eraser param
* Update image.py
rename rainbow eraser param
* Update KCC.ui
Rename rainbow eraser checkbox and tooltip
* Update KCC_ui.py
Rename erase rainbow checkbox and tooltip
* Update KCC_gui.py
Rename erase rainbow checkbox and option
* Update README.md
rename erase rainbow param
* Update KCC_gui.py
correct param name for eraserainbow
---------
Co-authored-by: Alex Xu <alexkurosakimh3@gmail.com>
* Save images in GIF when output is set to MOBI and forcepng option is used
* Save images in GIF when output is set to EPUB, kindle profile is used and forcepng option is set
* media-type="image/gif"
---------
Co-authored-by: Alex Xu <alexkurosakimh3@gmail.com>
* preserve margin
* set ratio to 0.5
* add preserve margin GUI
* increase step size to 5
* remove clear
* fix save
* save preserveMarginBox
* math
* max 99
* Eliminate unnecessary use of MD5 checksum
md5checksum() computes the actual checksum of a specified file, which is appropriately expensive, but the code seemed to be using the checksum result as a key into the imgMetadata dictionary to avoid handling image files being renamed during processing steps. This seems like a very expensive way to handle the rename so instead, I now update the imgMetadata keys with the new filename in the one place that the rename happens, and avoid MD5 checksums entirely.
* merge conflicts
* Add missing handling for image path renames due to nested chapter folder name
* merge conflicts
* merge conflicts
* add perf_counters
* imgFileProcessing perf_counter
* use startswith and removeprefix
---------
Co-authored-by: utopiafallen <utopiafallen@gmail.com>
Commit f952634971 moved image corruption detection out from the ComicPage constructor and into a standalone detectCorruption() function. This led to a performance regression because now corruption detection happens in a single thread when it used to be distributed across worker threads, and because a source image is now loaded twice in memory: once during corruption detection and once when actually going to process the image.
Image file corruption detection is now back inside the ComicPage constructor and the extra load() has been removed because the convert() call will automatically invoke load() and most likely throw the same exceptions.
* Inter-panel cropping method.
* 1. Save interpanelcrop option.
2. Update readme with the the new interpanelcrop argument.
3. Add a tooltip to the inter-panel crop box.
1. Replaced both crop margins and crop margins & page num with newer algorithm.
2. Crop max power level increased to 3.0
3. Adds NumPy as a new dependency.
