Shulou(Shulou.com)06/03 Report--

This article mainly introduces "how to use C++ programming template matching ultra-detailed recognition of handwritten digits", in daily operation, I believe many people in how to use C++ programming template matching ultra-detailed recognition of handwritten digits on the problem of doubt, small series consulted all kinds of information, sorted out simple and easy to use operation method, hope to answer "how to use C++ programming template matching ultra-detailed recognition of handwritten digits" doubts help! Next, please follow the small series to learn together!

The process is simple, as follows:

Matching success: There is a minimum distance (these distances are equal) and it is a number; there are multiple minimum distances and it is the same number.

Denial of recognition: There are multiple minimum distances, and they are different numbers.

Recognition error: There is a minimum distance, but it is not the same number as the measured number.

Perhaps at first glance do not understand, I explain here, understand can bypass. Let's assume that 1, 2, 3 (note that they all have multiple samples) are the training set, and d is the test sample. When matching, match to d with the smallest distance from 1 and only the smallest distance from 1,(may be the smallest distance from multiple 1 samples or only one) then match successfully; when matching to d with a sample of 1 and 2 have the smallest distance, then reject the match; when matching to d (if d is a sample of 1) and 2 have the smallest distance, then identify the error.

Because image processing is not the main topic of this article, we skip the image processing step (those interested can see the image processing course) and go directly to the processed images. So how do we build our training library, and how do we get computers to recognize our images? Next, let's look at how to build a training library.

In my experiment, I had 1000 training samples (200 test samples), and if I wanted the computer to recognize them, I had to digitize them. In the step of image processing, the training samples we get are all 28*28 pixel images, which can be thought of as a large number. In order to improve the processing speed, we compress the images into 7*7 size, which improves the processing speed and facilitates us to write code, because 7*7 and 4*4 are squares. As shown below:

Compressed images: We traverse the 7*7 grid vertically, counting the cells with more than 127 pixels, and when the number exceeds 6 (Some students think it should be 8, because 8 is half, but the correct rate of 8 is too low, so I found a suitable parameter) We will set the corresponding position of the 7*7 two-dimensional array corresponding to the large grid to 1, otherwise 0; and then convert the array to a string, so that we will get a string of length 49, which is the core of our computer matching.

In addition, I first put the training set and test set separately into data, and then take them out in turn for comparison. Can also be used while traversing the test set and processing, a comparison, I did not output the file name, because I used the way is relatively stupid, the amount of code is also a lot, mainly because I wrote before there are a lot of bugs, resulting in I can not run successfully, so I use this simple code to avoid errors, small partners do not have to use this way!

It is worth noting that the timing of file stream opening and closing will also greatly affect the code running, this problem has troubled me for a long time, I hope you take warning, the specific location in the code I have marked out. In addition, everyone knows about the file stream of reading files and writing files. Reading files is ofstream, writing files is ifstream, and every time you access files, you have to open files and close files. The getline function takes data one line at a time, so we don't close a document until we've traversed it, and we don't open it again.

Finally, to explain my string comparison, I used a flag rejected to indicate whether the current string was rejected when similarity was found. If the number is less than 49, it will be assigned to the similarity, and the rejection recognition will be set to false until the smallest one is found. When the smallest one is found and another with the same similarity is found, it will be judged whether the two sample numbers are the same. If not, the rejected will be set to true, that is, the rejection recognition will be directly output at the end.

I judge whether two samples are the same number by range comparison, which is simply 0--99 of the training sample corresponding to 0--19 of the test sample. This is a lazy method. I don't have time to change the code, so I replace the text with file name like others. (File name needs to be added when comparing file names)

Other explanations I put in the code, help everyone more direct understanding!

#include#include#include#include#include#include //api and structs #include#includeusing namespace std;using namespace cv;void ergodicTest(string filename, string name); //traversal function string Image_Compression(string imgpath); //compress image and return string int distance(string str1, string str2); //void compare();int turn(char a);void main(){ const char* filepath = "E:\\learn\\vsfile\\c++project\\pictureData\\train-images"; ergodicTest(filepath,"train_num.txt"); //Process training set const char* test_path= "E:\\learn\\vsfile\\c++project\\pictureData\\test-images"; ergodicTest(test_path, "test_num.txt"); compare(); } void ergodicTest(string filename,string name) //traverse and save path to files{ string firstfilename = filename + "\\*.bmp"; struct _finddata_t fileinfo; intptr_t handle; //cannot use long, because precision problems will lead to access conflicts,longlong can also be string rout = "E:\\learn\\vsfile\\c++project\\pictureData\\" + name; ofstream file; file.open(rout, ios::out); handle = _findfirst(firstfilename.c_str(), &fileinfo); if ( _findfirst(firstfilename.c_str(), &fileinfo) != -1) { do { file

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.