1
00:00:05,280 --> 00:00:08,650
We're going to start
experimenting with tesseract

2
00:00:08,650 --> 00:00:11,785
using just a simple image
of nice clean text.

3
00:00:11,785 --> 00:00:15,400
Let's first import PIL from
image and display the image

4
00:00:15,400 --> 00:00:19,945
text.png from PIL
will import image.

5
00:00:19,945 --> 00:00:22,720
Image is equal to image.open,

6
00:00:22,720 --> 00:00:25,700
and we'll get the text.png
other than the read_only.

7
00:00:25,700 --> 00:00:28,905
Then we'll display that image.

8
00:00:28,905 --> 00:00:33,265
Great, we have a base image
of some big clear text.

9
00:00:33,265 --> 00:00:36,910
Let's import pytesseract and
use the dir function to get

10
00:00:36,910 --> 00:00:38,170
a sense of what might be

11
00:00:38,170 --> 00:00:40,510
some interesting functions
to play with.

12
00:00:40,510 --> 00:00:42,765
So import pytesseract,

13
00:00:42,765 --> 00:00:46,100
and we can use dir to
see what's inside of it.

14
00:00:46,100 --> 00:00:48,800
Okay. It looks like
there is just a handful

15
00:00:48,800 --> 00:00:50,165
of interesting functions,

16
00:00:50,165 --> 00:00:53,645
and I think image_to_string
is probably our best bet.

17
00:00:53,645 --> 00:00:55,220
Let's use the help function to

18
00:00:55,220 --> 00:00:57,020
interrogate this a bit more.

19
00:00:57,020 --> 00:01:01,290
So help pytesseract
image_to_string.

20
00:01:03,070 --> 00:01:06,545
So this function takes an image
as the first parameter,

21
00:01:06,545 --> 00:01:08,510
and there is a bunch of
optional parameters,

22
00:01:08,510 --> 00:01:10,760
and it'll return
the results of the OCR.

23
00:01:10,760 --> 00:01:12,110
I think it's worth comparing

24
00:01:12,110 --> 00:01:13,790
this documentation string with

25
00:01:13,790 --> 00:01:15,050
the documentation we were

26
00:01:15,050 --> 00:01:16,760
receiving from the pillow module.

27
00:01:16,760 --> 00:01:20,540
Let's run the help command on
the image resize function.

28
00:01:20,540 --> 00:01:24,790
So help image, Image.resize.

29
00:01:25,610 --> 00:01:28,090
Notice how the
PILLOW function has

30
00:01:28,090 --> 00:01:29,860
a bit more information in it.

31
00:01:29,860 --> 00:01:31,270
First off, it's using

32
00:01:31,270 --> 00:01:34,430
a specific format called
reStructuredText,

33
00:01:34,430 --> 00:01:36,375
which is similar in intent

34
00:01:36,375 --> 00:01:38,840
to document markups such as HTML,

35
00:01:38,840 --> 00:01:40,330
the language of the web.

36
00:01:40,330 --> 00:01:41,950
The intent is to embed

37
00:01:41,950 --> 00:01:44,350
semantics in the
documentation itself.

38
00:01:44,350 --> 00:01:46,930
For instance, in
the resize function we see

39
00:01:46,930 --> 00:01:49,900
the words param size with
colon surrounding it.

40
00:01:49,900 --> 00:01:53,375
This allows documentation
engines which create docs for

41
00:01:53,375 --> 00:01:55,090
the source code to link

42
00:01:55,090 --> 00:01:58,270
the parameter to the extended
docs about that parameter.

43
00:01:58,270 --> 00:02:00,970
In this case, the extended
docs tell us that

44
00:02:00,970 --> 00:02:04,000
the size should be passed as
a tuple of width and height.

45
00:02:04,000 --> 00:02:07,455
Notice how the dogs for
image_to_string, for instance,

46
00:02:07,455 --> 00:02:09,530
indicate that there's
a lang parameter

47
00:02:09,530 --> 00:02:10,910
which we could use,

48
00:02:10,910 --> 00:02:12,890
but then fail to say
anything about what

49
00:02:12,890 --> 00:02:16,805
that parameter is for
or what its format is.

50
00:02:16,805 --> 00:02:20,450
What this really means is
that you need to dig deeper.

51
00:02:20,450 --> 00:02:21,950
Here's a quick hack
if you want to

52
00:02:21,950 --> 00:02:23,870
look at the source
code of a function.

53
00:02:23,870 --> 00:02:25,805
You can use the inspect gets

54
00:02:25,805 --> 00:02:28,475
source command and
print the results.

55
00:02:28,475 --> 00:02:30,350
So let's import inspect,

56
00:02:30,350 --> 00:02:32,000
and remember this
module comes from

57
00:02:32,000 --> 00:02:34,350
our Python three
standard library.

58
00:02:34,350 --> 00:02:36,760
Then we'll create the source

59
00:02:36,760 --> 00:02:39,830
inspect.getsource and we
pass it a function pointer.

60
00:02:39,830 --> 00:02:41,885
You note that we're not
calling the function.

61
00:02:41,885 --> 00:02:44,345
We're just passing
a reference to the function,

62
00:02:44,345 --> 00:02:47,770
and then let's print
that source to the screen.

63
00:02:47,770 --> 00:02:50,780
So it's interesting, you
can actually look at

64
00:02:50,780 --> 00:02:53,720
the source code behind
a given function,

65
00:02:53,720 --> 00:02:54,860
and that's one of the powers of

66
00:02:54,860 --> 00:02:57,635
an interpreted language
like Python.

67
00:02:57,635 --> 00:03:00,200
There's actually
another way in Jupyter,

68
00:03:00,200 --> 00:03:02,600
and that's to append
two question marks

69
00:03:02,600 --> 00:03:05,285
to the end of a given
function or module.

70
00:03:05,285 --> 00:03:07,410
Other editors have
similar features,

71
00:03:07,410 --> 00:03:09,920
and this is actually
a great reason that

72
00:03:09,920 --> 00:03:12,860
you should be using a software
development environment.

73
00:03:12,860 --> 00:03:15,040
So pytesseract.image_to_string

74
00:03:15,040 --> 00:03:16,130
just as if we were going to call

75
00:03:16,130 --> 00:03:18,170
the function add
two question marks

76
00:03:18,170 --> 00:03:19,550
to the end and then run that.

77
00:03:19,550 --> 00:03:23,690
We see that it pops
up at the bottom of

78
00:03:23,690 --> 00:03:25,850
the screen with a lot
more information and it's

79
00:03:25,850 --> 00:03:29,310
nice in syntax
highlighted for us too.

80
00:03:34,580 --> 00:03:37,270
We can see from the source
code that there

81
00:03:37,270 --> 00:03:39,430
really isn't much more
information about

82
00:03:39,430 --> 00:03:41,395
what the parameters are for or

83
00:03:41,395 --> 00:03:43,775
what this image_to_string
function is?

84
00:03:43,775 --> 00:03:45,300
This is because underneath,

85
00:03:45,300 --> 00:03:47,370
the pytesseract
library is calling

86
00:03:47,370 --> 00:03:50,685
a C++ library which does
all of the hard work,

87
00:03:50,685 --> 00:03:53,050
and the author just
passes through all of

88
00:03:53,050 --> 00:03:56,170
the calls to the underlying
tesseract executable.

89
00:03:56,170 --> 00:03:57,690
This is a common issue when

90
00:03:57,690 --> 00:03:59,355
working with Python libraries,

91
00:03:59,355 --> 00:04:01,870
and it means that we need
to do some web sleuthing

92
00:04:01,870 --> 00:04:03,340
in order to understand how we

93
00:04:03,340 --> 00:04:05,605
can interact with tesseract.

94
00:04:05,605 --> 00:04:07,490
In a case like this,

95
00:04:07,490 --> 00:04:10,495
I just Googled tesseract
command line parameters,

96
00:04:10,495 --> 00:04:13,135
and the first hit was
what I was looking for.

97
00:04:13,135 --> 00:04:16,160
Here's the URL to the GitHub.

98
00:04:17,890 --> 00:04:21,940
This goes to a wiki page
which describes how to call

99
00:04:21,940 --> 00:04:25,000
the tesseract executable,
and as we read down,

100
00:04:25,000 --> 00:04:27,670
we see that we can actually
have tesseract use

101
00:04:27,670 --> 00:04:30,445
multiple languages in
its detection such as

102
00:04:30,445 --> 00:04:33,145
English and Hindi
by passing them in

103
00:04:33,145 --> 00:04:37,240
as eng plus hin,
that's very cool.

104
00:04:38,570 --> 00:04:41,055
One last thing to mention,

105
00:04:41,055 --> 00:04:44,015
the image_to_string function
takes in an image,

106
00:04:44,015 --> 00:04:45,430
but the docs don't really

107
00:04:45,430 --> 00:04:47,875
describe what this image
is underneath.

108
00:04:47,875 --> 00:04:49,930
Is it a string to an image file?

109
00:04:49,930 --> 00:04:52,975
A PILLOW image or something else?

110
00:04:52,975 --> 00:04:55,630
Again we have to sleuth and or

111
00:04:55,630 --> 00:04:58,700
experiment to understand
what we should do.

112
00:04:58,700 --> 00:05:00,320
If we look at the source code

113
00:05:00,320 --> 00:05:01,850
for the pytesseract library,

114
00:05:01,850 --> 00:05:04,835
we see there's a function
called run_and_get_output.

115
00:05:04,835 --> 00:05:06,560
Here's a link to that function on

116
00:05:06,560 --> 00:05:08,840
the author's GitHub account.

117
00:05:08,840 --> 00:05:11,390
When we look at this function
we can actually

118
00:05:11,390 --> 00:05:16,110
see what actually happens
when we call this function.

119
00:05:16,340 --> 00:05:18,830
In this function,
we see that one of

120
00:05:18,830 --> 00:05:20,480
the first things which happens as

121
00:05:20,480 --> 00:05:23,870
the image is saved through
the save_image function.

122
00:05:23,870 --> 00:05:26,490
Here's that line of code.

123
00:05:29,060 --> 00:05:31,670
We see that there's
another function

124
00:05:31,670 --> 00:05:33,020
called prepare image,

125
00:05:33,020 --> 00:05:36,515
which actually loads the image
as a PILLOW image file.

126
00:05:36,515 --> 00:05:39,290
So, yes, sending a PIL image file

127
00:05:39,290 --> 00:05:41,600
is appropriate for use
for this function.

128
00:05:41,600 --> 00:05:43,250
It sure would have
been useful for

129
00:05:43,250 --> 00:05:45,440
the author to have included
this information and

130
00:05:45,440 --> 00:05:47,750
reStructuredText to help us

131
00:05:47,750 --> 00:05:50,825
not have to dig through
the implementation itself.

132
00:05:50,825 --> 00:05:53,075
But this is an open
source project.

133
00:05:53,075 --> 00:05:54,770
Maybe you would
like to contribute

134
00:05:54,770 --> 00:05:57,070
back some better documentation.

135
00:05:57,070 --> 00:05:59,145
Just a hint, if
you're interested in,

136
00:05:59,145 --> 00:06:02,055
the doc line we need
is param image,

137
00:06:02,055 --> 00:06:04,470
and then we just
say that it's a PIL

138
00:06:04,470 --> 00:06:08,265
Image.Image file or
an ndarray of bytes.

139
00:06:08,265 --> 00:06:10,880
In the end, we often don't

140
00:06:10,880 --> 00:06:12,920
do this full level
of investigation,

141
00:06:12,920 --> 00:06:14,960
and we just experiment
and try things.

142
00:06:14,960 --> 00:06:18,680
It seems pretty likely that a
PIL Image.Image would work,

143
00:06:18,680 --> 00:06:21,815
given how well known PIL
is in the Python world.

144
00:06:21,815 --> 00:06:24,260
But still, as you explore and use

145
00:06:24,260 --> 00:06:26,150
different libraries
you'll see a breadth

146
00:06:26,150 --> 00:06:28,190
of different documentation norms.

147
00:06:28,190 --> 00:06:31,505
So it's useful to know how
to explore the source code,

148
00:06:31,505 --> 00:06:33,485
and now that you're at
the end of this course,

149
00:06:33,485 --> 00:06:36,145
you've got the skills to do so.

150
00:06:36,145 --> 00:06:39,965
Okay. Let's try and run
tesseract on this image.

151
00:06:39,965 --> 00:06:41,210
So texts is equal to

152
00:06:41,210 --> 00:06:45,050
pytesseract.image_to_string
and we pass in the image,

153
00:06:45,050 --> 00:06:50,070
and then let's just print
out the text. Looks great.

154
00:06:50,070 --> 00:06:51,650
We can see that
the output includes

155
00:06:51,650 --> 00:06:55,099
new line characters and
faithfully represents the text,

156
00:06:55,099 --> 00:06:57,410
but it doesn't include
any special formatting.

157
00:06:57,410 --> 00:06:59,330
Let's go on and look at something

158
00:06:59,330 --> 00:07:02,280
with a bit more nuance
to it next.