Why am I getting blank output from TextRecognize for the following image?
Is it because of the size of the image?
Neither am I getting output for this code:
x = Import["https://i.stack.imgur.com/RbqrD.jpg"];
TextRecognize[x]
nor for this one:
x = Import["https://i.stack.imgur.com/RbqrD.jpg"];
img = ImagePartition[x, {50, 69}];
MatrixForm[img];
p=img[[4,1]];
TextRecognize[p]
It does not seem to recognise a single character in isolation. Even when I make it easier
y1 = ImageCrop[
ImageCrop[img1, {100, 225}, {Left, Bottom}], {20, 25}, {Left, Top}];
y2 = ImageResize[y1, Scaled[10]];
y3 = Sharpen[y2];
y4 = Binarize[y3]
We still have that
TextRecognize@y4
gives "" . But there is hope, see the next section.
Recognizing multiple random characters
TextRecognize can find characters that do not make up words (I'm not sure if sentences matter here). A key element here is that the characters are regularly spaced. Let's get an example from the docs
wolframImg =
ImageCrop[
ImageCrop[
ImageResize[
Import[
"http://reference.wolfram.com/language/ref/Files/TextRecognize.\
en/I_20.gif"]
,
Scaled[5]
]
,
{434, 240}
,
{Left, Bottom}
]
,
{408, Full}
,
{Right, Top}
]
![enter image description here][2]
TextRecognize works reasonably well on this, as advertised.
TextRecognize[wolframImg]
ZSI-IC HSKRN CHKRVD
TextRecognize actually often gives quite a few false positives, when it is confident about a lot of (other) characters. This gave me the (silly) idea to compose this image with characters from your example.
y4Copies = Table[y4, {2}, {5}];
y5 = ImageAssemble[y4Copies];
y6 = ImageResize[y5, First@ImageDimensions@wolframImg];
compoImg = ImageAssemble[{{y6}, {wolframImg}}]
Now we can do
TextRecognize[compoImg]
which gives
EEEEE EEEEE ZSHC HSKRN CHKRVD
Here is a shorter implementation of the idea of Jacob Akkerboom to assemble the characters in a line:
$Version
i = Import["https://i.stack.imgur.com/RbqrD.jpg"];
c = ComponentMeasurements[ColorNegate@Binarize@i, "Image", #"ConvexCoverage" < .9 &];
row = ImageAssemble[{ImagePad[ImageResize[#, {Automatic, 200}], 10] & /@ c[[;; , 2]]}]
TextRecognize[row]
"11.1.0 for Microsoft Windows (64-bit) (March 13, 2017)"
"ESEEEEEEEEEEEEEAAAAAAHHHHA"
The parameters 200 and 10 were carefully chosen by trial and error for achieving best possible recognition quality. And the latter is still leaves much to be desired. :(
Such poor quality of TextRecognize is especially surprising considering that Tesseract (which TextRecognize uses under the hood) shows excellent quality and does not require any assistance at all, as it can be checked for example using its Windows GUI gImageReader.
TextRecognize in version 11.1 make recognition of individual characters straightforward. But I see that it is not the case: RecognitionPrior -> "Character" improves nothing as well as the other option choices! It is a shame that WRI can't properly integrate Tesseract into Mathematica...
– Alexey Popkov
Apr 03 '17 at 07:52
TextRecognizeworks is that it looks for groups of letters and guesses what the word is. So it's better for complete sentences, it's not good for individual letters. In this image especially the letters have a spatial arrangementTextRecognizecannot make sense of, so you are better off usingImageCorrelateand some sample letters (there are other Q&As about this). – C. E. Jul 09 '14 at 09:55TextRecognizebeing used like this before, for example here, but it's just not very good at it. For varying spatial distributions it is a much safer bet to use image correlation. – C. E. Jul 09 '14 at 18:25TextRecogniseon one element of that matrix. You cannot expect that it somehow knows it is part of a matrix. Also the given partition does not tell you much about where the characters are, as well as that such behaviour ofTextRecognise(to work with matrices) is simply not implemented. – Jacob Akkerboom Jul 11 '14 at 08:36