Diary Archive (2005.3.4 ~ 2005.3.18)
Today, the JPEG standard (ITU T.81 is ISO / IEC 10918-1) is over. The impression is not very deep, it is very troublesome, it is very troublesome. Next, look at the JFIF document. Estimated To see the JPEG standard, and at the same time, it is necessary to see the implementation of the JPEG encoding decoding program. Test it, the CODEC written before, it takes 3 to 4 seconds to press JPEG. It is on the machine of CY 1.2G Measured. It used to write too much, and it is very irregular. To rewrite, add it to at least 0.5 seconds. The host should be able to do, the handheld is more troublesome, only 100m.ft.2005.3.4
I saw JFIF standard. I saw JPEG Codec Algorithm and Note, written in 1999. Feeling Encoder simple, and Decoder is more troublesome, there are more things to handle. His decoder is a bit messy. Basic processes can also understand. I don't understand the Scaled AA & N DCT, and the way I originally used. The color space conversion is not directly calculated according to the conversion matrix, and it is optimized. It is not very intuitive. Marking the MARKER, SYNTAX and Procedure used in Codec, Syntax and Procedure .2005.3.5
This morning, I went to the classroom. I started writing the opening report in the afternoon. I wrote the first part of the literature review. Column Several standards: H.261 Standard (ITU H.261) H.263 Standard (ITU H.263) JPEG Standard (ISO / IEC 10918 ITU T.81) MPEG Format MPEG-1 Standard (ISO / IEC 11172) MPEG-2 Standard (ISO / IEC 13818) MPEG-4 Standard (ISO / IEC JTCI / SC29 / WG11 N4668) painted A system block diagram. Talking on the paper ... 2005.3.6
Exhausted, write the opening report, I haven't worted it yet. Copy is also tired, I would rather do something substantive thing. Sweeping the color space conversion method of IJG Curia, the basic method is to understand. Just put the equation group The floating point ratio is expanded into an integer, avoiding floating point operations, and finally except (replacing the displacement, accelerate) with a proportional factor. But there are many details to prevent data overflow. 2005.3.7
Yesterday I finished the opening report. I saw the fdct and idct.fdct in IJG codec in the two days. I finally understood what I mean, and I used to use the formula. That is The factor is proposed. Finally quantify the multiplication without division. But theoretically did not explain the pass, I don't know how to put the S matrix. Hey, the linear algebra is not learned ... or check the information ... IDCT is really I don't understand, and I can't do it with my transposition matrix. But my flow map should be true, I have tried it. But the volume is definitely bigger than IJG: Source data: 140 144 147 140 140 155 179 179 144 152 140 147 140 148 167 167 163 162 152 172 168 145 156 160 152 155 136 160 162 148 156 148 140 136 147 162 147 167 140 155 155 140 136 162 136 156 123 167 162 144 140 147 148 155 136 155 152 147 147 136DCT (QValue = 100, the best quality, quantization step = 1) 187 -19 15 -10 24 -10 -14 -19 21 -35 27 -10 -10 10 14 7 - 10 -24 -1 5 -17 3 -20 -1 -8 -6 15 -15 -8 -3 -3 8 -3 9 9 1 -10 18 19 15 4 -3 -18 8 9 -4 1 -7 9 0 -3 3 -1 -8 -1 -2 0 -8 -2 1 1 4 -6 0 Reconstruction: 140 144 147 140 140 155 179 179 144 152 140 147 140 148 167 179 151 * 155 136 167 164 * 16 2 152 173 * 168 145 156 160 153 * 155 136 160 162 148 156 148 140 136 147 162 146 * 166 140 155 155 140 136 162 136 156 123 167 162 144 140 147 148 155 136 155 152 147 147 136 worth of * Different, it may be a floating point error. Not too clear. Like Double and Float. 2005.3.9 These two days have made some basic work. Learn and summarize the input output of several basic data types, do a few Experiment: DEV-C Laboratic DEV-C for several int type formats under Experiment DEV-C under the experimental DEV-C of several int type formats under the experimental DEV-C for several int type formats http: // blog .9cbs.net / jx_kingwei / find a lot of problems, don't understand ..., Juji ~~~~~~ 2005.3.11
After reading the matrix of a day, I dizzy. IDCT can be made. ~~~~~~~~~~~~~~~~~~~~~~~~~ 2005.3.14
FDCT / IDCT finally saw some doorways .... Both are the matrix fell down. Oh, although the iDCT and IJG reservoirs are different, but the algorithm complexity is almost ~~~~~~~~~~ ~~~ 2005.3.15
Today, I was asked to be stupid ... In the afternoon and evening, two modules were written: stream.cpp and ycbcr.cppijg Color space conversion method too metamorphosis ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Understand. Y = 0.29900 * R 0.58700 * G 0.11400 * b CB = -0.16874 * r - 0.33126 * G 0.50000 * B 128 CR = 0.50000 * R - 0.41869 * G - 0.08131 * B 128R = Y 1.40200 * (CR-128) G = Y - 0.34414 * (CB-128) - 0.71414 * (CR-128) B = Y 1.77200 * (CB-128) 2005.3.16 These two days in writing color spatial conversion and sub-sampling part For the three sampling formats in 411, 422, 444, 6 functions are written to complete the sub-sampling, and the upward sample process. It turns out that two functions are unified. Unfortunately, it is too much trouble, and the speed will slow down. Or separate, Use a function pointer to call. The original method of hopping points is too serious in the edge of the color block. To take average value. Now the effect is much more. Hehe ~~~~~ 2005.3.18
