|
From: Jakub S. <jst...@mu...> - 2017-05-13 23:19:00
|
Hello, Initially, I would like to thank for development of this library, I do appreciate the compression speed. Currently I'm using the "raw_data_in" mode and "jpeg_write_raw_data" function to encode YCbCr image. It works great when creating JFIF conformant output, but my current requirements is to produce raw "scan" content. Therefore, I have few questions about libjpeg-turbo API: 1. Is there any possibility to disable the writing of any markers and header segments? I managed to disable DQT and DHT tables, but my goal is to get Huffman encoded "scan" data only. Simply speaking I want libjpeg to perform transformation, DC prediction, entropy encoding and flush encoded MCUs into some buffer. The reason is to create RFC2435 conformant RTP payload, witch forces the transmission of some parameters in different way than JFIF. Obviously, I can strip unwanted markers and segments but it's a significant waste of time. 2. Same as 1. but encoding only a part of picture. I would like to divide the the work into multiple threads and at the end combine the content of output buffers (obviously with inclusion of restart markers). Currently, I'd have to provide fake image dimensions to the encoder (i.e. a quarter of height in case of using four threads) to by-pass the verification if entire picture have been encoded. Moreover, I'd have to perform above mentioned stripping of unwanted markers/segments. Unfortunately, I didn't found such a features in API exposed in "jpeglib.h" but maybe I missed something. -- Best regards Jakub Stankowski |
