I want to detect and extract texts in a natural image if it contains as google vision do. I found a library on GitHub that detect regions on image to find texts and after detection it does OCR. I want it to be faster and before text detection and extraction, I want to check if an image contains text or not.
I know I can run OCR on it but I want it to be faster than that. If it contains text then it should OCR, if not it should discard the image. Any ideas?
Related
I noticed that PNG files created by Gimp from the same RPG data are identical except for the very beginning. This image shows a diff of otherwise identical PNG files created with Gimp:
What is this data which changes each time and how is it encoded? Are there tools to decode it? Can you learn something from this information, e.g. can you find out when a PNG file was (probably) created by this information?
I was under the impression that PNG files are created deterministically* and don't store meta data which isn't necessary to decode the image. (Obviously, the last part is not true, either, as Gimp writes its own name into the files but doesn't ask the user (which is does if you export something as a JPEG file).)
* I use the word "deterministic" here to refer to things and only such which are the same on each execution/export/whatever given the same input. I'd usually use the word "functional" (i.e. like a mathematical function) but I fear this could be misunderstood by people who don't know what "functional" means in mathematics. Obviously, this is different from the usage of this word in information theory.
See the PNG header definition.
tIME stores the time that the image was last changed, so for me it's the same as the timestamp of the file you create.
bKGD gives the default background color. Possibly the bakcgournd color you are using in Gimp, or the color of the transparent pixels.
tEXT with key Comment and value Created with Gimp is just the default comment. You can change the comment for the image in Image>Properties and you can set a default comment in Edit>Preferences>Default Image
When I export the same PNG twice, I only see a change in tIME. In fact I can't get a bKGD item, even when exporting a PNG with transparent pixels. Are you using any specific options when exporting?
Can anyone suggest me how to convert a scanned image into a searchable image or a scanned pdf to a searchable pdf ?
I have been stuck in this situation since quite a while now.
i have tried pdfocr application in ubuntu but no success.
Tesseract version 3.03 supports creation of searchable PDF from image. For PDF, you can use GhostScript to convert it to image before sending it to Tesseract.
https://github.com/tesseract-ocr/tesseract
Currently, there is no right way of doing this on Ubuntu. All OCR engines output plain text and there is no way to add that text as a hidden layer on PDF over the image text.
Option 1: Use gscan2pdf which will make you a searchable PDF, but the OCRed text is placed in the top-left corner of the page, is invisible and much too small.
Option 2: Use PDF X-Change Viewer which has an option to OCR and works correctly by adding a text layer over the scanned image which is in concordance with it. You'll have to run it in wine, because it is a Windows application.
I am using matplotlib to draw a graph using some data and I have saved it in Pdf format.Now I want to add a logo to this file.How can I do this.
Thanks in advance
If you can do it the other way round, it is easier:
plot the image
load the logo from file with, e.g. Image module (PIL)
add the logo with plt.imshow, use the extent keyword to place it correctly
save the image into PDF
(You may even want to plot the logo first, so that it stays in the background.)
Unfortunately, this does not work with vector graphics, but as logos usually are not that large, you may use a .png or even a .jpg.
If you already have the PDF's then this is not a matplotlib or python question. You need some PDF editing tools or libraries to add the logo. Possible, but an entirely different thing.
I use Abbyy FineReader for ScanSnap to OCR a couple of scanned PDF files. The software claims it retains the original PDF images. The PDF file sizes pre-OCR and post-OCR are almost identical, which is good.
After the software is done, all PDF images appear anti-aliased in Acrobat X. Page navigation is much slower than before, and when I zoom in/out, the images first go to what looks like the pre-anti-aliasing version before quickly changing to anti-aliased images.
Left: Scanned PDF / Right: after OCR with Abbyy
I would like to get the original images without anti-aliasing back. Interestingly, when I open a single page from the anti-aliased PDF in Photoshop, there is no anti-aliasing and the image looks like the left one.
My limited PDF programming experience leads me to believe that Abbyy likely sets some kind of anti-alias flag for each image during OCR processing. How do I un-set this flag?
Any pointers to useful ideas would be much appreciated.
After the software is done, all PDF images appear anti-aliased in Acrobat X. Page navigation is much slower than before, and when I zoom in/out, the images first go to what looks like the pre-anti-aliasing version before quickly changing to anti-aliased images.
Actually in the original file 2013_11_15_22_51_31.pdf contains a JPEG image while the OCR'ed file 2013_11_15_22_51_31_OCR.pdf contains a JPEG2000 image.
Comparing them in third party viewers, it becomes clear that the image in the OCR'ed file is not inherently anti-alias'ed. Furthermore there is no evident flag in the PDF instructing PDF viewers to apply anti-aliasing to the JPEG2000 image. Thus, Adobe Reader seems to automatically render JPEG and JPEG2000 images differently, applying anti-aliasing to the latter but not to the former.
Comparing both images in detail, though, it becomes clear that these images are not identical but instead the image in the OCR'ed PDF is slightly rotated.
I assume Abbyy FineReader recognized that the original scanned image is not correctly oriented. Thus, it rotated it slightly to correct this orientation.
Thus, replacing the image in the OCR'ed version with the one from the original one is no option: Due to the rotation the OCR information would partially be somewhat off.
What you might want to try is to recode the JPEG2000 image to JPEG and replace the image in the OCR'ed version with this recoded one. This will mean some loss of quality but most likely you can get rid of the anti-aliasing this way.
Be aware, though, that the JPEG2000 image is slightly larger than the JPEG image to accomodate for the rotation.
PS: As #VadimR pointed out, there is indeed an /Interpolate true entry in the image dictionary of the OCR-ed version I missed when looking at the file. This does not seem to be the major issue slowing down the rendering.
There is /Interpolate true entry in image dictionary of OCR-ed version, and that's what causes 'anti-aliasing'. Whether that (and not JPEG2000 instead of JPEG compression) is a cause of slow-down, you check on large enough files.
To un-set this key, the best would be to turn it off while creating a file, and if that's not possible, to write and run a small program in suitable language.
But, since your file doesn't sport 'compressed objects' and offending key is in plain view inside a file, in the spirit of 'job done quickly' you can simply process your file e.g. like this:
perl -M-encoding -0777pe "s!/Interpolate true!' 'x17!ge" <in.pdf >out.pdf
I have an Autocad drawing which is a plan for land squares where each square contains a number.
I tried to convert it to image by choosing: File --> Export Data --> and file format Bitmap (bmp). (I have Autocad 2013 Mac version)
the file converted to image, but the quality is too bad, I can't see the land numbers inside the square when I zoom in the image.
I tried also with Postscript (PS file format), quality is a bit better but it's still bad.
Is there away to convert Autocad file to image but still preserve it's high quality details
I need to convert the file to image because I would like to publish it online on my website. maybe there other was to publish autocad file on line, if so, please advice. But the trick I want the background of the autcad (plan) to be transparent so that I could display it on top of Google maps. I if I used autocad plugin I can't make it transparent. right?
Use the PLOT (for the current drawing) or PUBLISH (for batching) commands to produce high quality images.