Cropping accuracy with Vision VNDetectRectanglesRequest

I am using VNDetectRectanglesRequest to detect rectangular images on a page, and crop them. Detection is quite accurate, but in most cases, the images aren't cropped clean. They have a thin white border around them, as old-fashioned photo prints used to have. Any suggestions? (I get much better cropping with OpenCV, but would like to use Vision if I can improve the cropping results.)

Have you tried using its properties such as minimumAspectRatio, maximumAspectRatio, minimumSize, regionOfInterest.
These are very useful (to me and had made alot of difference into my output) but still I'm also looking to make it more accurate.
Cropping accuracy with Vision VNDetectRectanglesRequest
 
 
Q