Primitive text direction feature detector #5
Conversation
| DEBUG = True | ||
|
|
||
| # Minimum number of detected text lines. | ||
| # If numer of lines recognized are below this value - the result is undefined |
|
Generally looks nice. Thank you for all your hard work |
| Algorithm is pretty straightforward and slow: | ||
|
|
||
| 1. increase shread contrast | ||
| 2. rotate image from -45 to +45 angles and compute |
|
|
||
| for i, value in enumerate(values): | ||
|
|
||
| if value > MAGIC_SECTIONS_THRESHOLD: |
There was a problem hiding this comment.
You only include into histograms for pixel[0]<MAGIC_COLOR_THRESHOLD (10) values 255-pixel[0]. So the minimum value in histogram would be 245, where it's not 0. I suggest replacing this check with "if value != 0:" or just "if value:" to make the code easier to follow
There was a problem hiding this comment.
Yeah, that is pretty dirty. Planned to remove that code and add image binarization in the beginning...
|
Guys, take a look into the latest version. :-) |
| # If number of lines recognized are below this value - the result is undefined | ||
| MIN_LINES_FOR_RESULT = 3 | ||
|
|
||
| # Magic values |
There was a problem hiding this comment.
The previous request still holds: please describe what is each of these "magic" values
|
By the way, in case of just a single line of text, your approach seems to mistake letters for separate lines, i.e. the answer is wrong by 90º. |
|
Guys, I've fixed code accoring your comments. |
|
|
||
| # All pixels above this value (after converting to grayscale mode) | ||
| # will be discarded (i.e. equal 255) | ||
| MAGIC_COLOR_THRESHOLD = 10 |
There was a problem hiding this comment.
Please remove MAGIC_ from here and other constants.
v.1.1 of text direction feature detector :)