-
Notifications
You must be signed in to change notification settings - Fork 0
/
NeuralNetwork.cpp
343 lines (304 loc) · 10.4 KB
/
NeuralNetwork.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
#include <iostream>
#include <opencv2/opencv.hpp>
#include <opencv2/core/core.hpp>
#include <fstream>
#include <chrono>
using namespace std;
using namespace cv;
Mat r2pTransform(Mat input)
{
Mat r2powImage, logImage;
//raise to power transform
input.convertTo(r2powImage, CV_32F); //changing pixel value type (from uchar to float)
r2powImage = r2powImage + 1; // dont know what is this :v
pow(r2powImage, 5, r2powImage);
normalize(r2powImage, r2powImage, 0, 255, NORM_MINMAX); //normalizing picture
convertScaleAbs(r2powImage, r2powImage); //converting back to uchar from float
//log transform
r2powImage.convertTo(logImage, CV_32F); //conversion of r2powImage from uchar to float
logImage = logImage + 1; //dont know what is this
log(logImage, logImage); //performing log transform
normalize(logImage, logImage, 0, 255, NORM_MINMAX); //normalizing picture
convertScaleAbs(logImage, logImage); //converting back to uchar
return logImage; //returning transformed image
}
Mat removeLeftBorder(Mat input)
{
Mat tmp = Mat(input.size().height, input.size().width - 2, input.type());
for (int i = 0; i < input.rows; i++)
{
for (int j = 2; j < input.cols; j++)
{
tmp.at<uchar>(i, j-2) = input.at<uchar>(i, j);
}
}
return tmp;
}
Mat SpreadRect(Rect r,Mat sourceImage)
{
Point tl = r.tl(); //top left corner Point
Point br = r.br(); //bottom right corner Point
int xIncrease = cvRound(r.width*0.15);
int yIncrease = cvRound(r.height*0.15);
if (tl.x - xIncrease > 0)
tl.x -= xIncrease;
else
tl.x = 0;
if (tl.y - yIncrease > 0)
tl.y -= yIncrease;
else
tl.y = 0;
if (br.x + xIncrease < sourceImage.cols)
br.x += xIncrease;
else
br.x = sourceImage.cols;
if (br.y + yIncrease < sourceImage.rows)
br.y += yIncrease;
else
br.y = sourceImage.rows;
Mat RectPart = sourceImage(Range(tl.y, br.y), Range(tl.x, br.x));
return RectPart;
}
//This functions finds bubbles on edges and user classifies those bubbles to correct or not correct using key input
//if marked area is bubble on edge press enter to calssify it as bubble on edge
//if marked area is not bubble on edge press 'n' to classify it as not bubble on edge
void prepareOutputFile(int argc, char** argv)
{
char key = 0;
ofstream outputFile;
outputFile.open("Out.txt");
for (int argP = 1; argP < argc; argP++)
{
Mat img = imread(argv[argP], IMREAD_GRAYSCALE);
Mat imgClean = imread(argv[argP], CV_LOAD_IMAGE_COLOR);
Mat imgCleanCopy = imgClean.clone();
Mat lbrImg = removeLeftBorder(img);
Mat o = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat poly = o.clone();
Point2f circleCenter;
float circleRadius;
Mat circleImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat contImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat singleContourImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat out;
vector<vector<Point>> contours;
vector<vector<Point>> circleContour;
vector<Point> polygonContour;
//some image preprocessing
Mat r2p = r2pTransform(lbrImg);
out = r2p.clone();
threshold(out, out, 100, 255, THRESH_BINARY);
Mat cannyOut;
Canny(out, cannyOut, 50, 240, 7);
findContours(cannyOut, contours, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE);
for (int i = 0; i < contours.size(); i++)
{
int contourPointOnEdgeCounter = 0;
if (contourArea(contours[i]) > 20)
{
for (int j = 0; j < contours[i].size(); j++)
{
if (contours[i][j].x == (lbrImg.cols - 1) || contours[i][j].x == 0 || contours[i][j].y == 0 || contours[i][j].y == (lbrImg.rows - 1))
{
contourPointOnEdgeCounter++;
if (contourPointOnEdgeCounter >= 1)
{
approxPolyDP(contours[i], polygonContour, 0.1, true);
minEnclosingCircle(polygonContour, circleCenter, circleRadius);
circle(singleContourImage, circleCenter, circleRadius, Scalar(255));
findContours(singleContourImage, circleContour, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE);
drawContours(imgClean, circleContour, -1, Scalar(0, 255, 0));
Rect r = minAreaRect(circleContour[0]).boundingRect();
Mat RectPart = SpreadRect(r, lbrImg);
resize(RectPart, RectPart, Size(60, 60));
rectangle(imgClean, r, Scalar(255, 0, 0));
imshow("Image with Circle", imgClean);
imshow("RectPart", RectPart);
key = waitKey();
outputFile << (int)circleCenter.x << " , " << (int)circleCenter.y << " , " << (int)circleRadius << " , ";
for (int k = 0; k < RectPart.rows; k++)
{
for (int l = 0; l < RectPart.cols; l++)
{
outputFile << (int)RectPart.at<uchar>(k, l) << " , ";
}
}
if (key == 13)
{
outputFile << "b" << endl;
}
if (key == 'n')
{
outputFile << "n" << endl;
}
break;
}
}
}
}
imgClean = imgCleanCopy.clone();
circleImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
contImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
singleContourImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
}
}
}
void testNeuralNetwork(int argc, char** argv, string annXMLName)
{
float annResult = 0;
int noOfFeatures = 3603;
Mat trainData = Mat(232, noOfFeatures, CV_32FC1, Scalar(0));
Mat trainClass = Mat(232, 2, CV_32FC1, Scalar(0));
Mat result = Mat(1, 2, CV_32FC1, Scalar(0));
Ptr<ml::ANN_MLP> ann = ml::ANN_MLP::create();
auto start = chrono::steady_clock::now();
ann = ml::ANN_MLP::load(annXMLName);
auto end = chrono::steady_clock::now();
if (!ann->isTrained())
{
cout << "Network not trained" << endl;
}
cout << "Loading NN from xml took: " << chrono::duration_cast<chrono::seconds>(end - start).count() << " sec" << endl;
for (int argP = 1; argP < argc; argP++)
{
Mat testSample = Mat(1, noOfFeatures, CV_32FC1, Scalar(0));
Mat img = imread(argv[argP], IMREAD_GRAYSCALE);
Mat imgClean = imread(argv[argP], CV_LOAD_IMAGE_COLOR);
Mat imgCleanCopy = imgClean.clone();
Mat lbrImg = removeLeftBorder(img);
Mat o = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat poly = o.clone();
Point2f circleCenter;
float circleRadius;
Mat circleImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat contImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat singleContourImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
Mat out;
vector<vector<Point>> contours;
vector<vector<Point>> circleContour;
vector<Point> polygonContour;
//some image preprocessing
Mat r2p = r2pTransform(lbrImg);
out = r2p.clone();
threshold(out, out, 100, 255, THRESH_BINARY);
Mat cannyOut;
Canny(out, cannyOut, 50, 240, 7);
findContours(cannyOut, contours, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE);
for (int i = 0; i < contours.size(); i++)
{
int contourPointOnEdgeCounter = 0;
if (contourArea(contours[i]) > 20)
{
for (int j = 0; j < contours[i].size(); j++)
{
if (contours[i][j].x == (lbrImg.cols - 1) || contours[i][j].x == 0 || contours[i][j].y == 0 || contours[i][j].y == (lbrImg.rows - 1))
{
contourPointOnEdgeCounter++;
if (contourPointOnEdgeCounter >= 1)
{
approxPolyDP(contours[i], polygonContour, 0.1, true);
minEnclosingCircle(polygonContour, circleCenter, circleRadius);
circle(singleContourImage, circleCenter, circleRadius, Scalar(255));
findContours(singleContourImage, circleContour, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE);
drawContours(imgClean, circleContour, -1, Scalar(0, 255, 0));
Rect r = minAreaRect(circleContour[0]).boundingRect();
Mat RectPart = SpreadRect(r, lbrImg);
resize(RectPart, RectPart, Size(60, 60));
rectangle(imgClean, r, Scalar(255, 0, 0));
imshow("Image with Circle", imgClean);
testSample.at<float>(0, 0) = (float)circleCenter.x;
testSample.at<float>(0, 1) = (float)circleCenter.y;
testSample.at<float>(0, 2) = (float)circleRadius;
int p = 3;
for (int k = 0; k < RectPart.rows; k++)
{
for (int l = 0; l < RectPart.cols; l++)
{
testSample.at<float>(0, p) = (float)RectPart.at<uchar>(k, l) / 255;
p++;
}
}
auto start = chrono::steady_clock::now();
annResult = ann->predict(testSample);
auto end = chrono::steady_clock::now();
cout << "Prediction time in seconds : " << chrono::duration_cast<chrono::milliseconds>(end - start).count() << " ms" << endl;
if (annResult == 0)
{
cout << "Bubble on edge" << endl;
}
waitKey();
break;
}
}
}
}
imgClean = imgCleanCopy.clone();
circleImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
contImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
singleContourImage = Mat(lbrImg.size(), lbrImg.type(), Scalar(0));
}
}
}
void trainNeuralNetwork()
{
ifstream inputFile;
inputFile.open("Out.txt");
if (!inputFile.is_open())
{
cout << "Cannot open file Out.txt" << endl;
}
string line, lineTmp, fileName;
int noOfFeatures = 3603;
Mat trainData = Mat(232, noOfFeatures, CV_32FC1, Scalar(0));
Mat trainClass = Mat(232, 2, CV_32FC1, Scalar(0));
int trainDataRowPointer = 0;
while (trainDataRowPointer < 232)
{
int i = 0;
for (i; i < 3603; i++)
{
inputFile >> line;
inputFile >> lineTmp;
if (i >= 3)
{
trainData.at<float>(trainDataRowPointer, i) = (float)stof(line) / 255;
}
else
{
trainData.at<float>(trainDataRowPointer, i) = (float)stof(line);
}
}
inputFile >> line;
if (line == "b")
{
trainClass.at<float>(trainDataRowPointer, 0) = 1.0;
}
else
{
trainClass.at<float>(trainDataRowPointer, 1) = 1.0;
}
trainDataRowPointer++;
}
Ptr<ml::ANN_MLP> ann = ml::ANN_MLP::create();
float lay[3] = { noOfFeatures,100,2 };
Mat layM = Mat(3, 1, CV_32F, lay);
ann->setLayerSizes(layM);
ann->setActivationFunction(ml::ANN_MLP::SIGMOID_SYM, 0.8, 0.8);
ann->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 300, 0.001));
ann->setTrainMethod(ml::ANN_MLP::BACKPROP, 0.1, 0.1);
cout << "Training started..." << endl;
auto start = chrono::steady_clock::now();
ann->train(trainData, ml::ROW_SAMPLE, trainClass);
auto end = chrono::steady_clock::now();
cout << "Training time in seconds : " << chrono::duration_cast<chrono::seconds>(end - start).count() << " sec" << endl;
if (!ann->isTrained())
{
cout << "Network not trained" << endl;
}
cout << "Network trained, saving network to ANN.xml" << endl;
ann->save("ANN.xml");
}
//int main(int argc, char** argv)
//{
// prepareOutputFile(argc, argv);
//}