
Libraries-notably Tesseract v5.3.0 which improves the OCR speed. Version is also compiled with the latest gcc compiler, v12.2.

See details in the k2pdfopt version history. I did some experimenting with Tesseract (OCR) v5.1 today. Has identical accuracy to Tesseract v4.1, but uses a new 32-bit floating point calculation It and benchmarked it with a standard test I have. Technique, that, interestingly, is 30% faster on the "best" english training file butĤ0% slower than Tesseract 4.1 on the "fast" english training file (on a Core i9-9900 CPU).
WINE FOR MAC OS X YOSEMITE DOWNLOAD
I've added a Linux Aarch64 binary to my download page which The optimum character height of a capital letter is between 25 and 35 pixels for the bestĪccuracy for both Tesseract v4.1 and v5.1.
WINE FOR MAC OS X YOSEMITE WINDOWS
I cross-compiled on a Debian 10 virtual linux box on my Windows PC. If anybody can give me feedback whether it works or not. I had a user request a binaryįor their Pinephone, so I'm hoping this will work. I've re-worked my download page a bit to try and make it smarterĪbout forcing a fresh load every time as opposed to the browser pulling it up fromĪn internal cache, which can cause problems with expired capcha values. The download page manually in your browser (click the refresh button). On the latest Macs with the M1 chip, which is a very impressive performer. See the tableīelow comparing k2pdfopt performance on a core i9-9900 vs. The "No OCR" row compares single-threaded an Apple M1 with twoĭifferent C compilers (I posted the clang v12 version).

Is not as dramatic, probably because Tesseract has optimizations for the hardware extensions in Performance since only the OCR processing in k2pdfopt is multithreaded. Interesting also that clang v12 beats gcc v11 handily.

Performance is even more impressive when you consider that its thermal design power (TDP) isĪbout 20 W compared to the i9-9900's 65 W. This version improves OCR multithreading, adds better DJVU support (text layer extraction),Īdds CBZ support, and is compiled with the latest third party libraries, e.g. This is primarily a bug-fix release, fixing over 20 issues that have accumulated over time.
