How i could make better this image for tesseract recognition?

[diegomage]

convert dIsuln.png -unsharp 0x10 A_blur_unsharp0.jpg
convert A_blur_unsharp0.jpg -negate -lat 10x10+10% -negate jx8MPc9_kuw2_lat10-0.png
convert jx8MPc9_kuw2_lat10-0.png -fill white -opaque "#00ff00" balloon_white0.gif
tesseract -psm 10 balloon_white0.gif scanned2.txt


I have a simple image but I dont know How i could make better



this command output an E instead 2

[fmw42]

The image is too small and too even colored to do OCR. There are limits to what you can do by way of character size in the image and how thick it is. Is it a 2 or a Z? How will OCR know, I a person cannot tell the difference. I doubt you can do anything to improve this to get the OCR to read it accurately. But I do not know that much about OCR. Perhaps someone else can advise

[diegomage]

But there is a way to remove that little point

[diegomage]

The small point on the upper left margin

[diegomage]

for output this

[diegomage]

My command works when there are no such small points that interferes in the image

[diegomage]

I see one of your command for filter blob but I dont know how use this
convert balloon_white0.gif -define connected-components:area-threshold=40 -connected-components 4 -threshold 0 jx8MPc9_kuw2_lat10_cc30.png

how I could use this Please help

[fmw42]

try

Code: Select all

convert 1FY60.png -channel red -separate +channel -threshold 70% result.png

[diegomage]

thanks ; but not works how expected however this command works better ;
convert balloon_white0.gif -colorspace gray -colors 2 +dither -type bilevel imageout.tif
convert imageout.tif -define connected-components:area-threshold=77 -connected-components -4 -auto-level hh

[diegomage]

but output a image with another colors

[fmw42]

Each image will likely be different and require different approaches. I do not know one universal solution.

[snibgo]

Input: 1FY60.png


This image has an object with a background that extends to the edges of the image. So we can remove the background by making transparent all pixels that are within a certain fuzz percentage of any edge pixel. For this image, 6% works fine.

Windows BAT syntax.

Code: Select all

call %PICTBAT%edgeColsTrans 1FY60.png 1fy_x1.png . 6

1fy_x1.png:

Code: Select all

%IM%convert ^
  1fy_x1.png ^
  -background White -layers Flatten ^
  -fill Black +opaque White ^
  -resize 300%% ^
  1fy_x2.png

1fy_x2.png:

Code: Select all

tesseract 1fy_x2.png 1fy_x2 -psm 10

This gives the expected result, "2".

[diegomage]

very thank you this help me so much