Tesseract OCR+OpenCV识别二进制壁纸照片效果不佳求C++改进方案

我正在开发一个个人项目，需要识别二进制壁纸的拍摄照片。目前用Tesseract OCR结合OpenCV能正常识别文本截图，但对拍摄的壁纸照片识别效果极差，输出存在不完整、错误的情况。希望在C++环境下获得改进建议，也可尝试其他库，最终要将识别结果解析转换为ASCII。

当前预处理及识别代码 #

#include <string>
#include <fstream>
#include <vector>
#include <iostream>
#include "basicOCR.h"


using namespace std;
using namespace cv;

void getText(string imPath, string outPath)
{
    char *outText;
    ofstream outputFile(outPath);

    double count, avg, lowest = 100;

    Mat im = cv::imread(imPath);
    if (im.empty())
    {
        cerr << "Error reading image file" << endl;
        return;
    }
    Mat imGray, imDenoised,imThresholded, imInverted;
    fastNlMeansDenoising(im, imDenoised);
    cvtColor(imDenoised, imGray, COLOR_BGR2GRAY);
    threshold(imGray, imThresholded, 0, 255, THRESH_BINARY_INV | THRESH_OTSU);
    // bitwise_not(imThresholded, imInverted);

    cv::imshow("idk", im);
    waitKey(0);
    cv::imshow("idk", imDenoised);
     waitKey(0);
    cv::imshow("idk", imGray);
    waitKey(0);
    cv::imshow("idk", imThresholded);
    waitKey(0);
    //cv::imshow("idk", imInverted);
   //waitKey(0);



    tesseract::TessBaseAPI *api = new tesseract::TessBaseAPI();

    if (api->Init(NULL, "eng", tesseract::OEM_LSTM_ONLY))
    {
        cerr << "Could not initialize tesseract " << endl;
        exit(1);
    }
    api->SetPageSegMode(tesseract::PSM_AUTO);
    api->SetVariable("tessedit_char_whitelist", "01 ");

    api->SetImage(imThresholded.data, imThresholded.cols, imThresholded.rows, imThresholded.channels(), imThresholded.step);
    api->Recognize(0);
    tesseract::ResultIterator *ri = api->GetIterator();
    tesseract::PageIteratorLevel level = tesseract::RIL_TEXTLINE;
    if (ri != 0)
    {
        do
        {
            const char *word = ri->GetUTF8Text(level);
            float conf = ri->Confidence(level);
            int x1, y1, x2, y2;
            ri->BoundingBox(level, &x1, &y1, &x2, &y2);
            
            // outputFile << word << endl;
            outputFile << word << "conft :" <<conf << "box: " << x1 << ", "<< y1 <<", "<< x2 <<", "<< y2<<endl;
            count++;
            avg += conf;
            if (conf < lowest)
            {
                lowest = conf;
            }

            delete[] word;
        } while (ri->Next(level));
    }
    avg = avg / count;
    //cout << "average: " << avg << " lowest: " << lowest << " count: " << count << endl;
    // Destroy used object and release memory
    api->End();
    delete api;
    delete[] outText;

    outputFile.close();

}

测试示例 #

拍摄的壁纸照片（识别失败） #

图片1

• 原图：
• 预处理后：
• 错误输出：

图片2

• 原图：
• 预处理后：
• 错误输出：

文本截图（识别成功） #

• 原图：
• 预处理后：
• 识别输出：

01010111011010000110010101110010011001010110000101110011001000000100100100101100001000000101010001101000011011
conft :88.0777box: 12, 12, 2008, 38
11011100110010111000100000010001110010111000100000010000110110110001100101011011010111001101101111011011100010110
conft :84.2204box: 12, 52, 2015, 78
...

该输出可正常转换为ASCII。

内容的提问来源于stack exchange，提问作者Macto24