Thomas Kaiser wrote: > Hello Anders > > Anders Blomdell wrote: >> Thomas Kaiser wrote: >>> Hello Anders >>> >>> Anders Blomdell wrote: >>>> Thomas Kaiser wrote: >>>> which indicates that both DQT and Huffman is present in the stream. >>> It is _not_ present in the stream. >> Then I don't understand where it comes from... > > I do, because I implemented it into gspca V1 ;-) > > The cam only streams the picture data. To code the stream a _static_ > JPEG header is used. The cam and the driver know this header. Therefor, > the driver just builds a valid JPEG image with the known header and the > stream from the cam. > >>> Anyway I did not found time to try with the JPEG header from PAC7311. >>> And I have to apologize because I did not tell you what work I did for >>> the PAC7311. See this: http://www.kaiser-linux.li/index.php?title=PAC7311 >>> >>> You should find the PAC7311 JPEG header in this source: >>> http://www.kaiser-linux.li/files/PAC7311/gspcav1-PAC7311-20070425.tar.gz >> Will look into it. > > Check decoder/gspcadecoder.c and search for pac7311_jpeg_header. > >>>>> Anyway PAC7311 is working AFAIK. >>>> Which doesn't contradict that it's encoded in the stream from the camera. >>> BTW, can you send me the header and some Bytes from your stream, the 02 >>> 8a 00 a2 80 28 a0 0a 28 thing. As ASCII in Email is OK. >> The entire (white?) image is in the attachment (QUALITY=0). > > I tried the PAC7311 header. See the result in the attached files :-( Thanks for trying though. I tested some more, and by starting at offset 7 (based on that ff00 indicates jpeg stream, and that byte 8 differs between quality 1 & 2, [byte 7 follows quality]) Size=152d, quality=1 ff d8 ff fe 28 3c 01 fc ff 00 45 66 9a 69 a2 95 ... Size=152a, quality=2 ff d8 ff fe 28 3c 02 f9 55 15 29 a5 52 95 34 a8 I get img0 with the DQT you found on your installation disk, while img1 is what you get with the standard DQT. Could it be that yuv=0,0,0 implies a black image? Length is still bogus though, and with yuv != 0 I still get garbage, so I assume there is some problem with the huffman coding still! I guess what I have to do, is to transform a yuv=0,0,0 image using the 17 different DQT tables you found and try to generate a huffman table that matches that data and the 17 different samples I have (of course this table will have unknowns, but it might be a starting point). /Anders -- Anders Blomdell Email: anders.blomdell@xxxxxxxxxxxxxx Department of Automatic Control Lund University Phone: +46 46 222 4625 P.O. Box 118 Fax: +46 46 138118 SE-221 00 Lund, Sweden
