📑   Data Building. Practice Project 1_0: Letter Detection

✒️   Code Modules & Settings

xxxxxxxxxx

 

1

#!python3 -m pip install tensorflow==2.6.0 \

2

#--user --quiet --no-warn-script-location

3

path='/home/sc_work/.sage/local/lib/python3.9/site-packages'

4

import sys,warnings; sys.path.append(path)

5

warnings.filterwarnings('ignore')

6

import os,zipfile,h5py,cv2,urllib,requests

7

import numpy as np,tensorflow as tf

8

file_path='https://raw.githubusercontent.com/'+\

9

          'OlgaBelitskaya/data/main/letters/'

10

img_size=int(2048)

11

file_pre,file_list,file_name='',[],''; lc,uc=0,0

12

tf_img,img,gray_img,edges=[],[],[],[]

13

names=[['lowercase','uppercase'],

14

       [s for s in u'абвгдеёжзийклмнопрстуфхцчшщъыьэюя'],

15

       ['single-colored paper','striped paper',

16

        'squared paper','graph paper']]

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   Image Loading

xxxxxxxxxx

 

1

def tf_get_img(file_name,file_path=file_path,img_size=img_size):

2

    input_file=urllib.request.urlopen(file_path+file_name)

3

    output_file=open(file_name,'wb')

4

    output_file.write(input_file.read())

5

    output_file.close(); input_file.close()

6

    img=tf.io.read_file(file_name)

7

    img=tf.image.decode_image(img,channels=3)

8

    img=tf.image.convert_image_dtype(img,tf.float32)

9

    shape=tf.cast(tf.shape(img)[:-int(1)],tf.float32)

10

    new_shape=tf.cast(shape*img_size/max(shape),tf.int32)

11

    img=tf.image.resize(img,new_shape)

12

    return img[tf.newaxis,:]

13

def cv_get_img(file_name,lower_canny,upper_canny,

14

               file_path=file_path,img_size=img_size):

15

    input_file=urllib.request.urlopen(file_path+file_name)

16

    output_file=open(file_name,'wb')

17

    output_file.write(input_file.read())

18

    output_file.close(); input_file.close()

19

    img=cv2.imread(file_name)   

20

    gray_img=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

21

    img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)

22

    edges=cv2.Canny(gray_img,int(lower_canny),int(upper_canny))

23

    cv2.waitKey(int(0))

24

    return img/255.,gray_img,edges

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

@interact(layout=dict(top=[['lower_upper','letter','background'],

2

                           ['lower_canny','upper_canny']]))

3

def _get_params(lower_upper=selector(names[0],default='uppercase'),

4

                letter=selector(names[1],default='б'),

5

                background=selector(names[2],

6

                                    default='single-colored paper'),

7

                lower_canny=selector([10,20,..,210],default=20),

8

                upper_canny=selector([10,20,..,210],default=120)):

9

    global file_pre,file_list,file_name,\

10

    tf_img,img,gray_img,edges,lc,uc    

11

    file_pre='%02d_'%names[0].index(lower_upper)

12

    file_pre+='%02d_'%names[1].index(letter)

13

    file_pre+='%02d_'%names[2].index(background)

14

    print('file prefix: '+file_pre)

15

    file_list=np.array([file_pre+'%02d'%(i+1)+'.jpg' 

16

                        for i in range(4)])

17

    file_name=file_list[0]

18

    lc,uc=lower_canny,upper_canny

19

    print('[lower_canny,upper_canny] = %s'%str([lc,uc]))

20

    tf_img=tf_get_img(file_name)

21

    img,gray_img,edges=cv_get_img(file_name,lc,uc)

22

    for el in [tf_img.numpy()[int(0)],img,gray_img,edges]:

23

        matrix_plot(el,figsize=(5,7)).show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   Contour Detection

xxxxxxxxxx

 

1

def get_closed_img(edges,kernel):

2

    kernel=cv2.getStructuringElement(

3

        cv2.MORPH_RECT,(int(kernel),int(kernel)))

4

    closed_edges=cv2.morphologyEx(edges,cv2.MORPH_CLOSE,kernel) 

5

    cv2.waitKey(int(0))

6

    return closed_edges

7

def get_contours(gray_img,closed_edges):

8

    (contours, _)=cv2.findContours(

9

        closed_edges.copy(),cv2.RETR_EXTERNAL,

10

        cv2.CHAIN_APPROX_SIMPLE)

11

    for c in contours: 

12

        arc=cv2.arcLength(c,True) 

13

        approx_pdp=cv2.approxPolyDP(c,float(.02)*arc,True) 

14

        contour_img=cv2.drawContours(

15

            gray_img,[approx_pdp],-int(1),

16

            (int(0),int(255),int(0)),int(2))

17

    cv2.waitKey(int(0))

18

    return contours,contour_img

19

kernel,edge_kernel,closed_edges,contours,contour_img=0,0,[],[],[]

20

@interact(layout=dict(top=[['select_kernel']]))

21

def _get_kernel(select_kernel=selector([3,5,..,17],default=7)):

22

    global kernel,edge_kernel,closed_edges,contours,contour_img

23

    edge_kernel=select_kernel

24

    closed_edges=get_closed_img(edges,edge_kernel)

25

    contours,contour_img=get_contours(gray_img,closed_edges)

26

    for el in [closed_edges,contour_img]:

27

         matrix_plot(el,figsize=(5,7)).show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   Data Building

xxxxxxxxxx

 

1

def get_imgs(contours,img_size_out=int(32)):

2

    imgs=[]

3

    for c in contours: 

4

        x,y,w,h=cv2.boundingRect(c)

5

        m=np.random.randint(int(7),int(21))

6

        if (w>int(30) and w<int(250) and h>int(30) and h<int(250)):

7

            y1=int(y*(int(1)-float(.0015)*m))

8

            y2=int((y+h)*(int(1)+float(.0005)*m))

9

            x1=int(x*(int(1)-float(.001)*m))

10

            x2=int((x+w)*(int(1)+.001*m))

11

            new_img=img[y1:y2,x1:x2]

12

            if min(new_img.shape[int(0)],

13

                   new_img.shape[int(1)])>int(30):

14

                new_img=cv2.resize(new_img,

15

                                   (img_size_out,img_size_out))

16

                imgs.append(new_img)

17

    return np.array(imgs)

18

imgs=get_imgs(contours); n_img=int(7)

19

@interact(layout=dict(top=[['image_number']]))

20

def _get_images(image_number=selector([2,3,..,10],default=7)):

21

    global imgs

22

    if len(imgs)<49: image_number=int(2)

23

    ga=[[matrix_plot(imgs[i+image_number*j]) 

24

         for i in range(image_number)] for j in range(image_number)]

25

    graphics_array(ga).show(frame=False)

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   Data ZIP Storing

xxxxxxxxxx

 

1

file_list_out=[]; file_name_zip=file_pre[:-1]+'.zip'

2

def get_zip(imgs,start=0):

3

    global file_name_zip,file_pre,file_list_out

4

    current_list_out=[]

5

    for i in range(len(imgs)):

6

        idx=start+i

7

        file_name_img=file_pre+'%04d'%idx+'.png'

8

        cv2.imwrite(file_name_img,imgs[i]*255.)

9

        if (file_name_zip not in os.listdir()):

10

            with zipfile.ZipFile(file_name_zip,'w') as f:

11

                f.write(file_name_img); f.close()

12

        else:

13

            with zipfile.ZipFile(file_name_zip,'a') as f:

14

                f.write(file_name_img); f.close()

15

        os.remove(file_name_img)

16

        current_list_out+=[file_name_img]

17

    file_list_out+=current_list_out 

18

    return file_list_out

19

file_list_out=get_zip(imgs)

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

with zipfile.ZipFile(file_name_zip) as z:

2

    with open(file_list_out[-1],'wb') as f:

3

        f.write(z.read(file_list_out[-1]))

4

example_img=cv2.imread(file_list_out[-1])   

5

matrix_plot(example_img/255,figsize=(3,3)).show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   Data H5 Storing

xxxxxxxxxx

 

1

file_name_h5=file_pre[:-1]+'.h5'

2

lbls=[[int(file_pre[:2]),int(file_pre[3:5]),int(file_pre[6:8])]

3

      for i in range(len(imgs))]

4

imgs=np.array(imgs,dtype='float32')

5

lbls=np.array(lbls,dtype='int32')

6

print(imgs.shape,lbls.shape)

7

maxlen=max([max([len(n) for n in names[i]]) 

8

            for i in range(len(names))])

9

nms=[np.array([np.string_(name.encode('utf-8')) 

10

               for name in names[i]],dtype='S%d'%maxlen) 

11

     for i in range(len(names))]

12

with h5py.File(file_name_h5,'w') as f:

13

    f.create_dataset('images',data=imgs,compression='gzip')

14

    f.create_dataset('labels',data=lbls,compression='gzip')

15

    for i in range(len(names)): 

16

        f.create_dataset('names%d'%(i+1),data=nms[i],

17

                         dtype='S%d'%maxlen,compression='gzip')

18

    f.close()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

with h5py.File(file_name_h5,'r') as f:

2

    keys=list(f.keys())

3

    print('file keys: '+', '.join(keys))

4

    images=np.array(f[keys[0]]); labels=np.array(f[keys[1]])

5

    names=[[el.decode('utf-8') for el in f[keys[i]]]

6

           for i in range(2,len(keys))]

7

    f.close()

8

ti=[names[i][labels[0,i]] for i in range(3)]

9

matrix_plot(images[0],title=str(ti),figsize=(3,3)).show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   ZIP Storing of All Images with the Same Labels

xxxxxxxxxx

 

1

if (file_name_zip in os.listdir()): os.remove(file_name_zip)

2

start=0; file_list_out=[]

3

letter_names=(

4

    [['name %d'%i,[np.string_(name.encode('utf-8')) 

5

      for name in names[i]]] for i in range(len(names))])

6

np.savetxt('letter_names.txt',[str(letter_names)],fmt='%s')

7

print('letter labels: '+letter_names[1][1][0].decode('utf-8')+\

8

      '-'+letter_names[1][1][32].decode('utf-8'))

9

with zipfile.ZipFile(file_name_zip,'w') as f:

10

    f.write('letter_names.txt')      

11

def file_exist(file_name,file_path=file_path):

12

    r=requests.head(file_path+file_name)

13

    exist={'True':'exists','False':'does not exist'}

14

    print(file_path+file_name,exist[str(r.status_code==int(200))])

15

    return r.status_code==int(200)

16

for f in file_list:

17

    if file_exist(f):

18

        img,gray_img,edges=cv_get_img(f,lc,uc)

19

        closed_edges=get_closed_img(edges,edge_kernel)

20

        contours,contour_img=get_contours(gray_img,closed_edges)

21

        imgs=get_imgs(contours)

22

        file_list_out=get_zip(imgs=imgs,start=start)

23

        start=int(file_list_out[-1][-7:-4])+1

24

    else: break

25

file_list_out[-1]

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

@interact(layout=dict(top=[['image_step']]))

2

def _get_examples(image_step=selector([10,20,..,100],default=10)):

3

    global file_name_zip,file_list_out

4

    with zipfile.ZipFile(file_name_zip) as z:

5

        for i in range(3):

6

            current_file_name=file_list_out[-1-image_step*i]

7

            with open(current_file_name,'wb') as f:

8

                f.write(z.read(current_file_name))

9

            example_img=cv2.imread(current_file_name)   

10

            matrix_plot(example_img/255,title=current_file_name,

11

                        figsize=(3,3)).show()

12

    z.close()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️   H5 Storing of All Images with the Same Labels

xxxxxxxxxx

 

1

file_name_h5=file_pre[:-1]+'.h5'; imgs,lbls=[],[]

2

def file_exist(file_name,file_path=file_path):

3

    r=requests.head(file_path+file_name)

4

    exist={'True':'exists','False':'does not exist'}

5

    print(file_path+file_name,exist[str(r.status_code==int(200))])

6

    return r.status_code==int(200)

7

for f in file_list:

8

    if file_exist(f):

9

        img,gray_img,edges=cv_get_img(f,lc,uc)

10

        closed_edges=get_closed_img(edges,edge_kernel)

11

        contours,contour_img=get_contours(gray_img,closed_edges)

12

        imgs0=get_imgs(contours); imgs.append(imgs0)

13

        lbls.append([[int(file_pre[:2]),int(file_pre[3:5]),

14

                      int(file_pre[6:8])]

15

                      for i in range(len(imgs0))])

16

imgs=np.array(np.vstack(imgs),dtype='float32')

17

lbls=np.array(np.vstack(lbls),dtype='int32')

18

print(imgs.shape,lbls.shape)

19

maxlen=max([max([len(n) for n in names[i]]) 

20

            for i in range(len(names))])

21

nms=[np.array([np.string_(name.encode('utf-8')) 

22

               for name in names[i]],

23

              dtype='S%d'%maxlen)

24

           for i in range(len(names))]

25

with h5py.File(file_name_h5,'w') as f:

26

    f.create_dataset('images',data=imgs,compression='gzip')

27

    f.create_dataset('labels',data=lbls,compression='gzip')

28

    for i in range(len(names)): 

29

        f.create_dataset('names%d'%(i+1),data=nms[i],

30

                         dtype='S%d'%maxlen,compression='gzip')

31

    f.close()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

with h5py.File(file_name_h5,'r') as f:

2

    keys=list(f.keys())

3

    print('file keys: '+', '.join(keys))

4

    images=np.array(f[keys[0]]); labels=np.array(f[keys[1]])

5

    names=[[el.decode('utf-8') for el in f[keys[i]]]

6

           for i in range(2,len(keys))]

7

    f.close()

8

randi=randint(0,images.shape[0]-1)

9

ti=[names[i][labels[randi,i]] for i in range(3)]

10

matrix_plot(images[randi],title='n: %d =>'%randi+str(ti),

11

            figsize=(3,3)).show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages