trainOCR: Extract a training dataset for OCR procedure


This function allows to extract a training dataset for OCR procedure performed by getExposure. It is currently optimized for stardot cameras.


trainOCR(image.path, nsamples=100)



The absolute path to a folder of JPEG binary images, as converted from RGB with the function binaryConvert()


The maximum number of sampled images to be used. No need to change it.


This function allows to prepare a training dataset that will be used in function getExposure(). You need to identify 0-9 numbers and the capital letter E (Exposure), which are then used in getExposure(). The procedure makes use of locator to subsequently crop your image, so make sure you know how this function works in your OS. When you run the function a first image pops up on your graphic device. You have to click with the mouse on topleft and bottom right of the rectangle you want to crop. I suggest to crop to the entire string of text with all picture information, so to have the largest sample of numbers in it. When you close locator (right-click in Linux-OS, but likely also GUI-dependent), the cropped image will show up providing a zoom to the selection. The title in the plot helps you to remember which numbers you still have to define. Choose a number (the order you choose numbers does not matter) and make a second crop around it (always topleft, bottomright). Close locator. A third zoom will show up, gridded pixel by pixel. Again, crop your number topleft bottomright with a rectangle that exactly includes all pixels of your number. Close locator. In R command line you will be asked to type the number you have just drawn, or letter E. Type the number and press Enter. You will be prompted to a new image where you follow the procedure again to identify other numbers. When you will ne done with all numbers and E letter, you will get a named list with 11 elements. Each element will be a binary matrix for each of your numbers, and letter E.


Gianluca Filippa <>

See Also


Questions? Problems? Suggestions? or email at

All documentation is copyright its authors; we didn't write any of that.