• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Subscribers 98 subscribers
  • Views 306 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Is it possible to have a « Partial decompression » output from TI jpeg decoder?

Sara Grassi
Intellectual 460 points

Dear Sirs, for our image recognition application we need to decompress jpeg files from the database, before doing matching with the query image.

Currently I am using the JPEG Decoder from TI,  Version 2.00.01.01, downloaded from here with a EVMDM6446 board and the codec engine framework, using the IIMGDEC1interface. Everything works fine, except that the decompression takes a bit to long for our application, and we are out of the specification (response time) for our product.  Our implementation is inefficient as we are first fully decompressing the image, then calculating DCT of the 8x8 blocks in order to do the matching (correlations) over this transformed data. So we could save much calculation time if we could have an output of the partial decompressed data.

So my question is: in the jpeg decoder library provided for TI, is there a means to have acces to partial decompressed data (8x8 quantized dct before reconstruction of the image)? This would be real useful for our product development.

Thanks and best regards,

Sara

  • 0
    TI__Genius 17215 points

    Sara,

    Unfortunately, The Jpeg encoder cannot output intermediate result. It is not a simple change due to design constraint and DM6446 is not an active product where we are doing any enhancements. If it is critical, you may like to get it touch with our 3P who can do it for you at some cost.

    regards

    Yashwant

  • 0 in reply to Yashwant Dutt
    TI__Genius 17215 points

    Sorry, i meant decoder and not encoder.

    regards

    YD

  • 0 in reply to Yashwant Dutt
    Intellectual 460 points

    Hello, thanks for your message and the useful information. Finally we managed to decrease the response time within specification by other means (pushing some processing, such as luminance extraction, from the ARM to DSP) so the immediate problem is solved, however I have an additional question:

    - If ever we need to optimize further we will have to write ourselves the (baseline) partial decompression. We will only need to implement variable length decoding [VLD] and zigzag scan [ZZ]. Do you know if there are C6x libraries containing these functions ?

    Thanks and best regards,

    Sara