오늘의 사건사고

컴퓨터비전 마지막 프로젝트로 만들었던 숫자인식기.
단순히 웹캠으로 숫자를 비춰주면 어떤 숫자인지 판별해서 출력해주는 프로그램이다.
전체 프로그램은 Python을 이용해 구현했고, ctypes-opencv를 이용해 영상 처리 부분을 구현했으며, GUI는 PyQT를 이용해서 구현했다.
시간이 좀 부족해서 단순히 기본 템플릿 이미지(0~9)와 입력받은 영상의 일치하는 정도를 본 다음 가장 일치하는 정도가 높은 숫자를 출력해주는 방식으로 구현했는데, 비교 할 때 지표를 좀 더 추가한다면 적중률이나 폰트에 따른 편차 등이 줄어들지 않을까 싶다.
Python만 이용해서 구현했더니 속도가 너무 떨어져서(초당 1프레임도 안 나오는 듯) 픽셀 연산 부분은 따로 C++을 이용해서 구현 한 다음, DLL로 만들어서 Python에서 해당 함수들을 호출하는 방식으로 구현했다.

- 소스코드

- 발표자료
- 실행 가능한 버전(압축을 푼 뒤 NumberRecognizer.exe 파일을 실행)

#include <stdio.h> /* * main.cpp * OpenCVApp * * Created by Seongwoo Lee on 11/11/08. * Copyright 2008 RyuiSaka. All rights reserved. * */ // This header includes all the opencv headers (cv, cxcore, highgui, ml) #include <OpenCV/OpenCV.h> int main(int argc, char** argv) { IplImage* sourceImage = cvLoadImage("./test.jpg", CV_LOAD_IMAGE_COLOR); IplImage* hsvImage = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 3); IplImage* hsvImage_h = cvCreateImage(cvGetSize(hsvImage), IPL_DEPTH_8U, 1); IplImage* hsvImage_s = cvCreateImage(cvGetSize(hsvImage), IPL_DEPTH_8U, 1); IplImage* hsvImage_v = cvCreateImage(cvGetSize(hsvImage), IPL_DEPTH_8U, 1); IplImage* intermediateImage = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); IplImage* morphologyIntermediateImage = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); IplImage* convertedImage = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 3); IplImage* morphologyImage = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 3); char* sourceWindow = "Source Image(BGR)"; char* convertedWindow = "Converted Image(applied to the threshold)"; char* morphologyWindow = "Converted Image2(applied to the threshold and morphology)"; cvNamedWindow(sourceWindow, CV_WINDOW_AUTOSIZE); cvNamedWindow(convertedWindow, CV_WINDOW_AUTOSIZE); cvNamedWindow(morphologyWindow, CV_WINDOW_AUTOSIZE); cvShowImage(sourceWindow, sourceImage); cvCvtColor(sourceImage, hsvImage, CV_BGR2HSV); cvSplit(hsvImage, hsvImage_h, hsvImage_s, hsvImage_v, NULL); cvCmpS(hsvImage_s, 40, intermediateImage, CV_CMP_GE); cvCopy(sourceImage, convertedImage, intermediateImage); cvShowImage(convertedWindow, convertedImage); cvErode(intermediateImage, morphologyIntermediateImage, NULL, 1); cvCopy(sourceImage, morphologyImage, morphologyIntermediateImage); cvShowImage(morphologyWindow, morphologyImage); cvWaitKey(0); return 0; }

선배들이 쉽다고 하는게 이해가 안갔는데, 막상 만들어놓고 보니 쉽긴 쉽다. - 프로그램 작동 동영상 /* * main.cpp * OpenCV Application * * Created by RyuiSaka on 08. 11. 09. * Copyright 2008 Kwangwoon University. All rights reserved. * */ #include <OpenCV/OpenCV.h> int main(int argc, char **argv) { IplImage* sourceImage = cvLoadImage("./test.jpg", CV_LOAD_IMAGE_COLOR); IplImage* convertedImage = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 3); IplImage* outputImage1 = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); IplImage* outputImage2 = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); IplImage* outputImage3 = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); char* sourceImageWindow = "Source Image Window"; char* outputWindowCh1 = "Output Image Window - Ch 1"; char* outputWindowCh2 = "Output Image Window - Ch 2"; char* outputWindowCh3 = "Output Image Window - Ch 3"; int key; cvNamedWindow(sourceImageWindow, CV_WINDOW_AUTOSIZE); cvShowImage(sourceImageWindow, sourceImage); while(1) { key = cvWaitKey(0); if(key == '0') break; switch (key) { case '1': cvSplit(sourceImage, outputImage1, outputImage2, outputImage3, NULL); cvNamedWindow(outputWindowCh1, CV_WINDOW_AUTOSIZE); cvShowImage(outputWindowCh1, outputImage1); cvNamedWindow(outputWindowCh2, CV_WINDOW_AUTOSIZE); cvShowImage(outputWindowCh2, outputImage2); cvNamedWindow(outputWindowCh3, CV_WINDOW_AUTOSIZE); cvShowImage(outputWindowCh3, outputImage3); break; case '2': cvCvtColor(sourceImage, convertedImage, CV_BGR2YCrCb); cvSplit(convertedImage, outputImage1, outputImage2, outputImage3, NULL); cvShowImage(outputWindowCh1, outputImage1); cvShowImage(outputWindowCh2, outputImage2); cvShowImage(outputWindowCh3, outputImage3); break; case '3': cvCvtColor(sourceImage, convertedImage, CV_BGR2HSV); cvSplit(convertedImage, outputImage1, outputImage2, outputImage3, NULL); cvShowImage(outputWindowCh1, outputImage1); cvShowImage(outputWindowCh2, outputImage2); cvShowImage(outputWindowCh3, outputImage3); break; case '4': cvCvtColor(sourceImage, convertedImage, CV_BGR2HLS); cvSplit(convertedImage, outputImage1, outputImage2, outputImage3, NULL); cvShowImage(outputWindowCh1, outputImage1); cvShowImage(outputWindowCh2, outputImage2); cvShowImage(outputWindowCh3, outputImage3); break; case '5': cvCvtColor(sourceImage, convertedImage, CV_BGR2XYZ); cvSplit(convertedImage, outputImage1, outputImage2, outputImage3, NULL); cvShowImage(outputWindowCh1, outputImage1); cvShowImage(outputWindowCh2, outputImage2); cvShowImage(outputWindowCh3, outputImage3); break; case '6': cvCvtColor(sourceImage, convertedImage, CV_BGR2Lab); cvSplit(convertedImage, outputImage1, outputImage2, outputImage3, NULL); cvShowImage(outputWindowCh1, outputImage1); cvShowImage(outputWindowCh2, outputImage2); cvShowImage(outputWindowCh3, outputImage3); break; case '7': cvCvtColor(sourceImage, convertedImage, CV_BGR2Luv); cvSplit(convertedImage, outputImage1, outputImage2, outputImage3, NULL); cvShowImage(outputWindowCh1, outputImage1); cvShowImage(outputWindowCh2, outputImage2); cvShowImage(outputWindowCh3, outputImage3); break; default: break; } } return 0; }

우여곡절 끝에 겨우겨우 첫번째 Assignment를 끝냈다. /* * main.cpp * OpenCVApp * * Created by Seongwoo Lee on 11/03/08. * Copyright 2008 RyuiSaka. All rights reserved. * */// This header includes all the opencv headers (cv, cxcore, highgui, ml) #include <math.h>#include <stdlib.h>#include <string.h>#include <OpenCV/OpenCV.h>void cropWindow(int event, int x, int y, int flags, void* sourceImage);bool flag = FALSE;CvPoint savedPoint;int main(int argc, char **argv){ IplImage* sourceImage = cvLoadImage("./test.jpg", CV_LOAD_IMAGE_COLOR); char* window1 = "Test Window"; cvNamedWindow(window1, CV_WINDOW_AUTOSIZE); cvShowImage(window1, sourceImage); cvSetMouseCallback(window1, cropWindow, sourceImage); cvWaitKey(0); cvReleaseImage(&sourceImage); cvDestroyAllWindows(); return 0 ;}void cropWindow(int event, int x, int y, int flags, void* sourceImage){ CvFont font; double hScale = 1; double vScale = 1; int lineWidth = 1; cvInitFont(&font, CV_FONT_HERSHEY_SIMPLEX, hScale, vScale, 0, lineWidth); if(event == CV_EVENT_LBUTTONDOWN && flag == FALSE) { savedPoint.x = x; savedPoint.y = y; flag = TRUE; } else if(event == CV_EVENT_LBUTTONDOWN && flag == TRUE) { int i, j; double r, g, b, avg; char* cropWindow = "Crop Window"; if(savedPoint.x > x) { int temp = savedPoint.x; savedPoint.x = x; x = temp; } if(savedPoint.y > y) { int temp = savedPoint.y; savedPoint.y = y; y = temp; } int diffX = abs(x-savedPoint.x); int diffY = abs(y-savedPoint.y); cvNamedWindow(cropWindow, CV_WINDOW_AUTOSIZE); IplImage* croppedImage = cvCreateImage(cvSize(diffX, diffY), IPL_DEPTH_8U, 1); IplImage* src_red = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); IplImage* src_green = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); IplImage* src_blue = cvCreateImage(cvGetSize(sourceImage), IPL_DEPTH_8U, 1); cvSplit(sourceImage, src_blue, src_green, src_red, NULL); for(i = 0; i < diffY; i++) { for(j = 0; j < diffX; j++) { r = cvGetReal2D(src_red, savedPoint.y+i, savedPoint.x+j); g = cvGetReal2D(src_green, savedPoint.y+i, savedPoint.x+j); b = cvGetReal2D(src_blue, savedPoint.y+i, savedPoint.x+j); avg = (r+g+b)/3; cvSetReal2D(croppedImage, i, j, avg); } } cvShowImage(cropWindow, croppedImage); cvReleaseImage(&croppedImage); flag = FALSE; }}