hi i resized the image and it change the dpi of the image from 300 dpi to 90 dpi . What is the correct way to resize image without changing its dpi in ocpencv . if we feed the 90 dpi image into OCR it barley recognizes any text in the image .
opencv does not deal at all with dpi, which is irrelevant in computer-vision and only needed for printing out things.
imread() will discard all meta information, and imwrite() won't write any dpi values.
the value of 90 you see (where ?) in your imageviewer is plain bogus / default (because there never was anything)
and again, neither ocr nor cnns care about dpi, only about pixels and image size.
The value 90 i see it in image properties by right clicking the image ->properties->Details->Horizontal resolution . how to maintain the pixel value consistent in the resized image ? the pixel value too decreases after resizing making the text in the images bit distorted how to avoid this ?
Four things: 1 When you shoot an object, the photo will have a virtual DPI - or actually PPI - related to that object. When you resize the photo that value necessarily changes. 2 OCR doesn't care about DPI. It only cares about having sufficient number of pixels to represent the letter, "Pixels Per Letterform", that changes when you resize the photo. 3 DPI values that any application writes to values are arbitrary and symbolic. They have (almost always) no relevance in image processing. Only pixels count. 4 So, if you don't have enough pixels after resizing, you must not resize.
so how do we resize the image in opencv such that the Pixels Per Letterform is not affected ? where can we see this value of Pixels Per Letterform ? im not able to understand the 4th point , resizing the image is the most common task in text detection and OCR how can we work without resizing ?
im going to resize the image in opencv so that i can CNN on this resized image to detect text features .You cannot run it on the orginal image size as CNN used stride to check every region in image , so we need to resize it and it will give the text coordinates which i will pass to ocr to crop and get text . the problem is when i resize the image text is very hard to recognize when compared to the original image .
The value 90 i see it in image properties by right clicking the image ->properties->Details->Horizontal resolution . how to maintain the pixel value consistent in the resized image ? the pixel value too decreases after resizing making the text in the images bit distorted how to avoid this ?
Asked: 2020-01-03 06:53:09 -0600
Seen: 3,976 times
Last updated: Jan 03 '20
