识别然后储存成文件，再用c语言调用Windows的MIDI的api，c语言没有map真的难受。别的不多说，上代码和视频。
#Numbered musical notation recognization
import cv2
from skimage.metrics import structural_similarity as ssim
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
import os
import numpy as np
import imutils
#training_path = "D:\\NMN\\train\\"
def reading_img_train(training_path,n,x,y):
""" # arg: k people in total, and each person has n photos
# The path of the parameter orl
# This function returns a list of all photos
# Read the folders in the order of first the name and then the serial number"""
imagesArray_gray = np.zeros((n*12, x*y))
i=-3
print("start reading")
while i <= 8:
print(i,"th")
for j in range(n):
full_path = training_path + str(i) + '//' + str(j+1) + '.jpg'
img = cv2.imread(full_path)
img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
_, threshold_img = cv2.threshold(img_gray, 120, 255, cv2.THRESH_BINARY_INV)
char_resized = cv2.resize(threshold_img, (x, y))
img_Vector = np.reshape(char_resized,(1,x*y))
imagesArray_gray[((i+3)*n)+j,:]=(img_Vector)
i=i+1
print("already read training photos")
return imagesArray_gray
def reading_img(img_path):
"""
This function reads img file then strengthen the picture
args:
img_path: you should provide file path here
The return value threshold_img,img_shape,origin img
"""
img = cv2.imread(img_path)
gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, threshold_img = cv2.threshold(gray_img, 120, 1, cv2.THRESH_BINARY_INV)
img_shape=threshold_img.shape
return threshold_img,img_shape,img
def project_to_y(threshold_img,img_shape,rate_bar,rate_hight):
"""
This function project picture to y axis
args:
threshold_img
img_shape
return edge_inf_y,row
"""
row=0
img_hight=img_shape[0]
img_width=img_shape[1]
bar_width=int(rate_bar*img_width)
char_hight=int(rate_hight*img_hight)
threshold_img_arr = np.uint8(np.array(threshold_img))
#project the image to y-axis
y_projection=np.sum(threshold_img_arr,axis=1)
"""print("inf of y")
for i in range(len(y_projection)):
print(y_projection[i])"""
#find the horizontal line(area)
in_num_flag=0
edge_temp=0
edge_inf_y=[]
for i in range(img_hight):
if in_num_flag==0 and y_projection[i]>bar_width:
in_num_flag=1
edge_temp=i
elif in_num_flag==1 and y_projection[i]<bar_width and (i - edge_temp) >char_hight:
in_num_flag=0
#if (i - edge_temp) >50,10:
edge_inf_y.append([edge_temp,i+int(char_hight*0.2)])
row=row+1
if (i*2 - edge_temp)<img_hight:
i=i*2 - edge_temp
print("projected y")
return edge_inf_y,row
def project_to_x(threshold_img,img_shape,edge_inf_y,row):
"""
This function project picture to x axis
args:
threshold_img
img_shape
return the full inf about edges
"""
img_hight=img_shape[0]
img_width=img_shape[1]
x_projection=[]
col_info=[]
#first we need to enhance image again
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
dilate_img = cv2.dilate(threshold_img, kernel)
#find the number row by row
edge_inf=[]
for i in range(row):
cropImg= dilate_img[edge_inf_y[i][0]:edge_inf_y[i][1], 0:img_width]
crophight,cropwidth=cropImg.shape
cropImg_arr=np.uint8(np.array(cropImg))
#project the image to x-axis
x_projection=np.sum(cropImg_arr,axis=0)
#find the vertical line(area)
in_num_flag=0
edge_temp=0
each_col=0
for j in range(img_width):
if in_num_flag==0 and x_projection[j]>(int(crophight*0.20)) and x_projection[j]<(int(crophight*0.68)):
in_num_flag=1
edge_temp=j
elif in_num_flag==1 and x_projection[j]<(int(crophight*0.25)) and (j - edge_temp) >(int(cropwidth*0.01)):
in_num_flag=0
edge_inf.append([edge_inf_y[i][0],edge_inf_y[i][1],edge_temp,j+1])
each_col=each_col+1
col_info.append(each_col)
print("projected x")
return edge_inf,col_info
def ncc(test_char_cmprs,train_char_cmprs):
"""# NCC classifier
# arg: test_char_cmprs,train_char_cmprs
# Return value: the name label of the closest number(list)"""
print("doing ncc")
row_test,_= np.shape(test_char_cmprs)
row_train,_= np.shape(train_char_cmprs)
test = float("inf")
result=[]
flag=0
for j in range(row_test):
test = float("inf")
for i in range(row_train):
dist_test = np.sqrt(np.sum(np.square(test_char_cmprs[j,:] - train_char_cmprs[i,:])))
if dist_test < test:
test = dist_test
flag=i
result.append((flag//60)-3)
print("ncc done")
return result
def gen_pca_field(edge_inf,img,numbers,train_path,x,y,r):
"""
generate two 2-d arrays(reduced dimension,to r)
"numbers" is the number of training pictures
x and y present the size of pictures,
only if all pictures are same size we can train the data
"""
#crop pictures and arrange them into 2-d arrary
print("doing pca")
gray_img_whole = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, threshold_img = cv2.threshold(gray_img_whole, 100, 255, cv2.THRESH_BINARY_INV)
test_char=np.zeros((len(edge_inf), x*y))
for i in range(len(edge_inf)):
temp_img=threshold_img[edge_inf[i][0]:edge_inf[i][1], edge_inf[i][2]:edge_inf[i][3]]
char_rcgnz=cv2.resize(temp_img,(x, y))
img_Vector = np.reshape(char_rcgnz,(1,x*y))
test_char[i,:]=(img_Vector)
print("croping done")
#then read the train pictures and arrange them into 2-d arrary
train_char=np.zeros((numbers*12, x*y))
train_char=reading_img_train(train_path,numbers,x,y)
print("reading train done")
"""for i in range(numbers):
train_read = cv2.imread(str(train_path)+str(i)+".jpg")
gray_img = cv2.cvtColor(train_read, cv2.COLOR_BGR2GRAY)
_, threshold_img = cv2.threshold(gray_img, 120, 255, cv2.THRESH_BINARY_INV)
char_rcgnz=cv2.resize(threshold_img,(x, y))
img_Vector = np.reshape(char_rcgnz,(1,x*y))
train_char[i,:]=(img_Vector)"""
#do pca
pca = PCA(n_components=r)
print("generated pca")
pca.fit(train_char)
print("training pca")
test_char_cmprs = pca.transform(test_char)
print("transforming")
train_char_cmprs = pca.transform(train_char)
print("pca done")
return test_char_cmprs,train_char_cmprs
if __name__ == '__main__':
img_path="D:\\NMN\\NMN7.jpg"
#img_path="/home/range/Code/pythonProject4/jianpu3.jpg"
#img_path="D:\\pythonProject4\\xunlian\\xunlian4.jpg"
#save_path="D:\\NMN\\save\\"
train_path="D:\\NMN\\train\\"
#read picture
threshold_img,img_shape,img=reading_img(img_path)
print("This is img size")
print(img_shape)
#project to y&x axis
edge_inf_y,row=project_to_y(threshold_img,img_shape,0.001,0.018)
edge_inf,col_info=project_to_x(threshold_img,img_shape,edge_inf_y,row)
#import recognized characters into files
#expo_obj(edge_inf,img,save_path)
#Do pca
test_char_cmprs,train_char_cmprs=gen_pca_field(edge_inf,img,60,train_path,61,165,10)
#result
result = ncc(test_char_cmprs,train_char_cmprs)
mylog = open('D:\\NMN\\mylog.txt', mode = 'a',encoding='utf-8')
count=0
for i in range(row):
#temp = cv2.imread(str(save_path)+str(result[i])+".jpg")
#cv2.imshow(str(i),temp)
for j in range(col_info[i]):
print(result[count],end=' ',file=mylog)
count=count+1
print("\n", file=mylog)
mylog.close()
#show
rectangle_img=img
for i in range(len(edge_inf)):
rectangle_img = cv2.rectangle(rectangle_img, (edge_inf[i][2], edge_inf[i][0]), (edge_inf[i][3], edge_inf[i][1]),(255, 0, 0), thickness=1)
cv2.namedWindow('Rectangle Image', 0)
#cv2.resizeWindow('Rectangle Image', 600, 500)
cv2.imshow("Rectangle Image", rectangle_img)
os.system("start D:\\vss1\\ConsoleApplication1\\x64\\Debug\\ConsoleApplication1.exe")
cv2.waitKey(0)
#include <stdio.h>
#include <stdlib.h>
#include <Windows.h>
#pragma comment(lib,"winmm.lib")
#pragma warning(disable:4996)
enum Scale//这是群里的大仙帮我找到的音阶参数
{
Rest = 0, C8 = 108, B7 = 107, A7s = 106, A7 = 105, G7s = 104, G7 = 103, F7s = 102, F7 = 101, E7 = 100,
D7s = 99, D7 = 98, C7s = 97, C7 = 96, B6 = 95, A6s = 94, A6 = 93, G6s = 92, G6 = 91, F6s = 90, F6 = 89,
E6 = 88, D6s = 87, D6 = 86, C6s = 85, C6 = 84, B5 = 83, A5s = 82, A5 = 81, G5s = 80, G5 = 79, F5s = 78,
F5 = 77, E5 = 76, D5s = 75, D5 = 74, C5s = 73, C5 = 72, B4 = 71, A4s = 70, A4 = 69, G4s = 68, G4 = 67,
F4s = 66, F4 = 65, E4 = 64, D4s = 63, D4 = 62, C4s = 61, C4 = 60, B3 = 59, A3s = 58, A3 = 57, G3s = 56,
G3 = 55, F3s = 54, F3 = 53, E3 = 52, D3s = 51, D3 = 50, C3s = 49, C3 = 48, B2 = 47, A2s = 46, A2 = 45,
G2s = 44, G2 = 43, F2s = 42, F2 = 41, E2 = 40, D2s = 39, D2 = 38, C2s = 37, C2 = 36, B1 = 35, A1s = 34,
A1 = 33, G1s = 32, G1 = 31, F1s = 30, F1 = 29, E1 = 28, D1s = 27, D1 = 26, C1s = 25, C1 = 24, B0 = 23,
A0s = 22, A0 = 21
};
enum Voice
{
L1 = C3, L2 = D3, L3 = E3, L4 = F3, L5 = G3, L6 = A3, L7 = B3,
M1 = C4, M2 = D4, M3 = E4, M4 = F4, M5 = G4, M6 = A4, M7 = B4,
H1 = C5, H2 = D5, H3 = E5, H4 = F5, H5 = G5, H6 = A5, H7 = B5,
LOW_SPEED = 500, MIDDLE_SPEED = 400, HIGH_SPEED = 300,
_ = 0XFF
};
void play()
{
FILE* fp = NULL;
int read_char=0;
fp = fopen("D:\\NMN\\mylog.txt", "r");
HMIDIOUT handle;
midiOutOpen(&handle, 0, 0, 0, CALLBACK_NULL);
int volume = 0x7f;
int voice = 0x0;
int sleep = 500;
int tmp;
while (!feof(fp))
{
fscanf(fp, "%d", &read_char);
printf("%d\n", read_char);
switch (read_char)
{
case (-7):
tmp = C3;
break;
case (-6):
tmp = D3;
break;
case (-5):
tmp = E3;
break;
case (-4):
tmp = F3;
break;
case (-3):
tmp = G3;
break;
case (-2):
tmp = A3;
break;
case (-1):
tmp = B3;
break;
case (0):
tmp = 0XFF;
break;
case (1):
tmp = C4;
break;
case (2):
tmp = D4;
break;
case (3):
tmp = E4;
break;
case (4):
tmp = F4;
break;
case (5):
tmp = G4;
break;
case (6):
tmp = A4;
break;
case (7):
tmp = B4;
break;
case (8):
tmp = C5;
break;
case (9):
tmp = D5;
break;
case (10):
tmp = E5;
break;
case (11):
tmp = F5;
break;
case (12):
tmp = A5;
break;
case (13):
tmp = B5;
break;
default:
tmp = 0XFF;
break;
}
voice = (volume << 16) + (tmp << 8) + 0x94;
midiOutShortMsg(handle, voice);
if (tmp != 0XFF)
Sleep(sleep);
else
Sleep(50);
}
midiOutClose(handle);
fclose(fp);
}
int main()
{
play();
return 0;
}
视频如下
方法还是比较笨，然后c语言写的不简练，识别还需要加强，不过好在是出效果了，继续优化。
Views: 121