I tried applying tesseract ocr on image but before applying OCR I want to improve the quality of the image so that OCR efficiency increase,
How to detect image brightness and increase or decrease the brightness of the image as per requirement.
How to detect image sharpness
It's not easy way to do that, if your images are similar to each other you can define "correct" brightness and adjust it on unprocessed images.
But what is "correct" brightness? You can use histogram to match this. See figure bellow. If you establish your correct histogram you can calibrate other images to it.
Richard Szeliski, Computer Vision Algorithms and Applications
I think you can use the autofocus method. You must check contrast histogram of image and define what is sharp to you.
source
*Basic recommendation, using the gray images is better on OCR
You can use the BRISQUE image quality assessment for scoring the image quality, it's available as a library. check it out, a smaller score means good quality.
I have the following problem:
I am receiving various scanned PDF files from a Kyocera Scanner Device.
I have to automatically manipulate these PDF Files in order to:
Delete the colors from textmarkers
Convert the PDF to grayscale
Put it in our DMS
I am using a Bash-Script to do the job.
For deleting the textmarker colors and converting to grayscale I use Imagemagick:
convert -density 150 INPUT.pdf \
-channel rgba \
-alpha set \
-fuzz 15% \
-fill white \
-opaque 'rgb(255,200,195)' \
-opaque 'rgb(255,253,177)' \
-opaque 'rgb(255,155,240)' \
-opaque 'rgb(255,91,193)' \
-colorspace gray OUTPUT-convert.pdf
The resulting image is quite good, BUT the size of the PDF is huge:
Original: 365K
Converted: 1.358K
So I've found a ghostscript command to do the job and reduce the file size:
gs -dSAFER -dBATCH -dNOPAUSE -dNOCACHE -dCompatibilityLevel=1.4 \
-sDEVICE=pdfwrite \
-sColorConversionStrategy=/LeaveColorUnchanged \
-dPDFSETTINGS=/ebook \
-sOutputFile=OUTPUT-ghostscript.pdf OUTPUT-convert.pdf
Now the file sizes are:
Original: 365K
Converted: 1.358K (OUTPUT-convert.pdf)
Ghostscript: 500K (OUTPUT-ghostscript.pdf)
I can't figure out why the size of the PDF after manipulation, from color to grayscale, is higher then the original document. The density (150 dpi) is the resolution of the original document.
When I put the converted PDF (1.358K) through Adobe Acrobat on Windows and recreate the PDF, the size is 213K. I have no loss in quality. How can I achieve this under linux with a bash script?
Any help is appreciated!
Here is a link for example PDF Files:
http://62.75.158.162/download/yKLu3fkbLy7MgkczDrKdG6osHdXh3jvy/
Its not really possible to comment very much without seeing an example file, to determine exactly what has happened at each stage.
However, I very strongly suspect that you have 'lost quality', its just that, at screen resolutions, you can't tell. Your original PDF file was created using ImageMagick at a resolution of 150 dpi. Most probably the image is stored uncompressed in the PDF file, which is why its large.
When you run that PDF file back through Ghostscript there are two effects. Firstly you've used the PDFSETTINGS canned set of job configuration. That (amongst many other things) downsamples grey images to a resolution of 150 dpi (so fortunately for you, no effect). It also compresses the image data using JPEG compression.
Now I've no idea what's in the original PDF file, but if the data there was compressed using JPEG, as seems likely, then you are double applying JPEG quantisation. That's a lossy process and will result in a loss of quality.
Since you are altering the original image data (to change the colour) you have no choice about decompressing the image data. However, to preserve quality you should then not use JPEG compression again, instead you should use Flate compression. The compression ration won't be as good, but it will keep the quality unchanged. To do that you would need to specify the GrayImageFilter using distillerparams, you can't use a PDFSETTINGS for that.
I can't imagine what Acrobat has done to decrease the file size still further (and you haven't said how you 'recreate the PDF file'), but I would imagine it involves reducing the quality of the image still further. Its hard to see how it could save 50% of the file size without doing so. Its also possible it is (like Ghostscript) JPEG compressing the grayscale data but using a more aggressive set of JPEG parameters (resulting in still more loss of quality, of course).
If you posted examples of the original, Ghostscript output, and Acrobat output I might be able to tell you more, but not from this.
For what its worth, there's a new feature in Ghostscript (requires version 9.23 or better) which allows you to create a PDF file which consists only of an image, and choose the colour model. You could run the original PDF file through Ghostscript using something like:
gs -sDEVICE=pdfimage8 -r150 -sOutputFile=gs.pdf
which would produce a pretty minimal PDF file where the original input has been rendered to a gray scale image (at 150 dpi), and that image wrapped up as a PDF file. I've no idea if that might work better for you.
Later EDIT
Yep, its pretty much what I expected.
The original file has what appears to be marked JPEG compression artefacts (all the rectangular 'speckles' round the text). Obviously without seeing the original document I can't tell whether this is because the original document was a JPEG printed to paper, or whether the artefacts were introduced by the scanner, or (more likely) whatever application converted the scanned image into a PDF. Checking the image stored in the PDF file I see that it is indeed a JPEG image.
Nevertheless, the original image is (in my opinion) really very noisy.
Now the output from 'convert' is arguably slightly better (in terms of legibility) than the original. I presume this is 'something' to do with your convert command line, can't be sure. The image in this case is not a JPEG, its compressed with RunLength encoding which is of course lossless. Its also less efficient as a compression method, so the image is bigger. For reasons best known to ImageMagick it also applies a soft mask to the image data. So that's two images per page now instead of just 1. Not too surprising that its larger than the original!
I suspect that the soft mask is due to your command line including RGBA. I assume that produces an alpha channel, and PDF doens't support simple alpha channel blending, its own transparency model is much more sophisticated. So I sort of suspect you are actually making the output file here larger than it needs to be. I'm afraid I can't help you with ImageMagick, I don't know anything about it, but getting rid of that second image would help a great deal.
Note that both your original file and the output from ImageMagick are essentially uncompressed (in terms of the PDF file 'structure').
Then we come to the Ghostscript produced PDF. The 'structure' of the PDF file is itself compressed, giving small size benefits. The images are all JPEG compressed, giving additional compression, but at the cost of quality. Applying JPEG quantisation multiple times always costs quality. By simply comparing the output from 'convert' with the output from Ghostscript I can easily see the degradation in quality.
Now we come to the Acrobat output. Ccomparing it with the other files it shows the worst quality. The JPEG artefacts are very clearly visible in the displayed image. In this case both the image and the soft mask have been compressed with the JPEG2000 compression scheme, which is a 'better' compression than JPEG. However, it looks like applying it to data which has already been quantised for JPEG yields pretty poor quality results. Or at least, applying it to a soft-masked JPEG image does :-)
The main problem with JPEG2000 is that it is patent encumbered. While decoders can be written royalty-free, to write an encoder you must licence the patented technology from the (many) patent holders, an expensive process.
So the AGPL version of Ghostscript does not include a JPEG2000 decoder, and as such cannot write JPEG2000 images.
Obviously you could use a copy of Acrobat to rewrite your PDF file with JPEG2000 compression as you have done here.
Assuming you want to avoid doing that, then my suggestion would be to investigate why convert is producing an image with a soft mask applied. I strongly suspect this is due to the use of rgba instead of rgb.
Avoiding the creation of the second (soft mask) image would (I believe) significantly decrease the size of the PDF file produced by 'convert'. You could gain at least some additional benefit, without any loss of quality, by running it through Ghostscript's pdfwrite device and specifying /FlateEncode for the GrayImageFilter. That would produce a PDF file where the PDF furniture is compressed, and where a better compression scheme is applied to the image data.
You could also just leave the Ghostscript line as it is, the quality degradation may be enough for you to live with.
if you use ubuntu you can try this on the command line. the result is impressive
Install ghostscript, for Ubuntu/Debian:
sudo apt-get install ghostscript
Resize your pdf with the command:
gs -sDEVICE=pdfwrite -dCompatibilityLevel=1.4 -dPDFSETTINGS=/screen -dNOPAUSE -dQUIET -dBATCH -sOutputFile=output.pdf input.pdf
Replace the file names output.pdf and input.pdf with your file names.
PDFs can start as vectors. But once you read it into ImageMagick, it gets rasterized. When writing back to PDF, it just imbeds the raster image into a vector PDF shell. So it has not been re-vectorized.
Your use of -density 150 has increased the rasterized file. The nominal density is 72. So have right there increased by 4x, which would just about cover your size increase. I think you stated your increase wrong. It probably should be Original: 365K Converted: 1.358M not Original: 365K Converted: 1.358K
Also if the scanned PDF was a raster in a vector, it may have had limited colors in palette form or simply compressed JPG form. Your rasterizing has converted to 24-bit color and by processing has increased the colors. So even as non-compressed grayscale it is larger.
You can compress your output PDF in ImageMagick as follows by writing the raster image to compressed JPG format and piping to another convert to write to PDF.
convert -density XXX input.pdf ... -colorspace gray -quality 50 JPG:- | convert - output.pdf
Adjust the quality value as desired
I am using the following command to generate progressive JPEGs using mozjpeg (cjpeg utility)
cjpeg -quality 85 -outfile outputfile.jpg inputfile.jpg
On rendering this output image, a grayscale image is first shown which is then followed by a greenish image and then the original image. After quite a bit of study, i found that splitting of DC coefficients is done to reduce the size of image. But, i need the rendering to be a blurred to bright one. How to achieve this?
There's an option for this:
-dc-scan-opt DC scan optimization mode
- 0 One scan for all components
- 1 One scan per component (default)
- 2 Optimize between one scan for all components and one scan for 1st component
plus one scan for remaining components
The default causes grayscale look, because it sends luma separately from chroma.
So I am trying to extract text from image. And as the quality and size of image is not good, it is giving inaccurate results. I tried few enhancements and other things with PIL but that is only worsening the quality of image.
Can someone suggest some enhancement in image to get better results. Few Examples of images:
In the provided example of image the text is visually of quite good quality, so the question is how it comes that OCR gives inaccurate results?
To illustrate the conclusions given in further text of this answer let's run the the given image
through Tesseract. Below the result of Tesseract OCR:
"fhpgearedmomrs©gmachom"
Now let's resize the image four times and apply thresholding to it. I have done the resizing and thresholding manually in Gimp, but with appropriate resizing method and threshold value for PIL it can be for sure automated, so that after the enhancement you get an image similar to the enhanced image I have got:
The improved image run through Tesseract OCR gives following text:
"fhpgearedmotors©gmail.com"
This demonstrates that enlarging an image can help to achieve 100% accuracy on the provided text-image example.
It may appear weird that enlarging an image helps to achieve better OCR accuracy, BUT ... OCR was developed to convert scans of printed media to texts and expect 300 dpi images of the text by design. This explains why some OCR programs didn't resize the text by themselves to improve their results and do bad on small fonts expecting higher dpi resolution of the image which can be achieved by enlarging.
Here an excerpt from Tesseract FAQ on github.com prooving the statement above:
[There is a minimum text size for reasonable accuracy. You have to consider resolution as well as point size. Accuracy drops off below 10pt x 300dpi, rapidly below 8pt x 300dpi. A quick check is to count the pixels of the x-height of your characters. (X-height is the height of the lower case x.) At 10pt x 300dpi x-heights are typically about 20 pixels, although this can vary dramatically from font to font. Below an x-height of 10 pixels, you have very little chance of accurate results, and below about 8 pixels, most of the text will be "noise removed".]
My tiff pyramid image conversion failed with the following error:
JPEGPreEncode: Strip/tile too large for JPEG
I had been using the following command:
vips tiffsave 1000x512.v newHUGEfile.tiff --compression jpeg --vips-progress --vips-cache-trace --bigtiff
Is there a way to avoid this symptom ?
How large is your test image? The jpeg encoder can't go over 65536 pixels in any axis. If your image is wider than that, you can't write it as a jpeg-compressed strip image.
Try adding --tile. This writes a tiled tiff image, where each tile is a separate 256x256 pixel region. Now the jpeg compressor will never see huge chunks at once and it ought to work.