by Zheng Qiushi
This week , I have done three tasks as follows:
我要评论 ① Review the process of encoding and decoding
② Analyze the procedure of dectecting and decoding in opencv.
③ Finish steps before the data infomation decoding
In this report , I’m going to focus on the last two points.
We can run the QR-code detector by calling the funtion " detectAndDecode( )" in modules/objdetect/src/qrcode.cpp .
Here is a flowchat of how it works.
In detect( ) , this function is implemented by counting the exact pixels in the image and trying to match the Position Detection Pattern (abbreviated to PDP in this report ) .
In my opinion , this function can be optimized, because in reality , the proportion of the PDP in the picture will be distorted by the perspective bending .So the direct counting method does not work and the detection rate is pretty low.
Maybe we can use the method of finding the coutours.According to the contour level , we can find the contour of three layers and select it as a candidate.
Because my main task is to finish the decoder ,so I will do this optimization later.
In decode( )
The main function is fullDecodingProcess( ),which calls other four procedures.
In updatePerspective( ): Correcting the image by doing an affine transformation.
In versionDefinition( ): Estimate the version roughly by doing a floodFill method,
In samplingForVersion( ):Sampling the pixels into a data array .First, enlarging the image .And then colleting the pixels in the smallest unit of tile which is a n*n matrix.
In decodingProcess( ) : Applied the third party Library "quirc".
My main task falls on the decodingProcess( ).
Instead of using a uint16 to store data , I select Mat as the container,the length is equal to its cols.
Mat mat_format(1,FORMAT_LENGTH,CV_8UC1,Scalar(0));
And the data can be easily get by calling
mat_format.ptr(0)[i]=...
In correct_format( ), I prefer to using hamming distance to correct the format infomation .
int hamming_detect(uint16_t fmt){
int best_fmt = -1;
int best_dist = 15;
int test_dist;
int index=0;
for(;index<32;index++){
uint16_t test_code=after_mask_format[index];
test_dist=hamming_weight(fmt ^ test_code);
/*find the smallest distance*/
if (test_dist < best_dist){
best_dist = test_dist;
best_fmt = index;
}
/*can't match two components*/
else if(test_dist == best_dist) {
best_fmt = -1;
}
}
cout<<"best_fmt : "<<best_fmt<<" best_dist : "<<best_dist<<endl;
return best_fmt;
}
Because , the data string is only 5bits (2bits for ECC level and 3bits for mask pattern) with a fixed ECC generator ,so there are only 32 components in the look-up table ,which can be quickly matched. We don’t need to do a “berlekamp_massey” method as it used to .
static const uint16_t after_mask_format [32]={
0x5412,0x5125,0x5e7c,0x5b4b,0x45f9, 0x40ce,0x4f97,0x4aa0,0x77c4,0x72f3,
0x7daa,0x789d,0x662f,0x6318,0x6c41, 0x6976,0x1689,0x13be,0x1ce7,0x19d0,
0x0762,0x0255,0x0d0c,0x083b,0x355f, 0x3068,0x3f31,0x3a06,0x24b4,0x2183,
0x2eda,0x2bed
};
Besides I also complete some functions to do the encode task.
The generator can be implement by doing multiplication in GF.
g4(x) = (x - α0) (x - α1) (x - α2) (x - α3)
Mat rs_generator_poly(int nsym) {
/*initialize the first item*/
Mat g(1,nsym+1,CV_8UC1,Scalar(0));
g.ptr(0)[0]=1;
g.ptr(0)[1]=1;
for(int i =1;i<nsym;i++){
/*get n as item (x - α0)*/
Mat n(1,2,CV_8UC1,Scalar(0));
n.ptr(0)[0]=gf_pow(2, i); n.ptr(0)[1]=1;
Mat r=gf_poly_mul(g, n);
for(int j=0;j<r.cols;j++){
g.ptr(0)[j]=r.ptr(0)[j];
}
}
cout<<"rs_generator_poly : "<<show_poly(g)<<endl;
return g;
}
This function is for getting the ECC for the data string ,which is implemented by doing a poly division.
/* 12 da df
* -----------------------
*01 0f 36 78 40 ) 12 34 56 00 00 00 00
* ^ 12 ee 2b 23 f4
* -------------------------
* da 7d 23 f4 00
* ^ da a2 85 79 84
* ---------------------
* df a6 8d 84 00
* ^ df 91 6b fc d9
* -------------------
* 37 e6 78 d9
*/
The divident is the data string ,the divisor is the generator and the remainder is the ECC.
Mat gf_poly_div(const Mat& dividend,const Mat& divisor) {
/* Note that the processing starts from the item with high number of times, so item [total-i] is processed for the i-th round */
int times=dividend.cols-(divisor.cols-1);
int dividend_len=dividend.cols-1;
int divisor_len=divisor.cols-1;
/*Mat.ptr(0)[i] stores the coeffient of the x^i*/
Mat r=dividend.clone();
for(int i =0;i<times;i++){
uint16_t coef=r.ptr(0)[dividend_len-i];
if(coef!=0){
for (int j = 0; j < divisor.cols; ++j) {
if(divisor.ptr(0)[divisor_len-j]!=0){
r.ptr(0)[dividend_len-i-j]^=gf_mul(divisor.ptr(0)[divisor_len-j], coef);
}
}
}
}
Mat ecc=r(Range(0,1),Range(0,10)).clone();
return ecc;
}
2020.07.11
本文地址：https://blog.csdn.net/weixin_43851149/article/details/107279354
如对本文有疑问， 点击进行留言回复！！
InnoDB体系架构（后台线程和内存池）（读MySQL技术内幕-InnoDB存储引擎）
.Net Core +EF Core+Log4Net基本三层架构搭建
网友评论