When i try to recognize a text in image, like the italian word "Perchè", Vision API get back the word "Perche" (give back the "e" and not the correct one "è").
I don't want to use languageHints to try to obtain better results because i've to do OCR Recognition across different language.
What is the problem here?
This is known issue with the Cloud Vision API when you don't use language hints.
You can see the actual bug report here.
It is in state accepted, but there seems to be radio silence on it for the last few months. It may take some time to roll it out.