 |
 |
My understanding is that a flatbed or sheetfed document scanner that produces 300 dpi will produce much better OCR results than a cheap digital camera that produces 300 dpi. The reasons have to do with the resolution and distortion of the resulting image, where resolution is defined as the number of line pairs per mm can be resolved (for example when scanning a test chart) - in other words the details that will show up for character images, and distortion is image aberration that can appear at the edges of the page image areas, particularly when illumination is not even. A scanner has much more even illumination. At 11:21 AM 5/1/2009 -0700, Erik Hetzner wrote: >At Fri, 1 May 2009 09:51:19 -0500, >Amanda P wrote: > > > > "On the other hand, there are projects like bkrpr [2] and [3], > > home-brew scanning stations build for marginally more than the cost of > > a pair of $100 cameras." > > > > Cameras around $100 dollars are very low quality. You could get no where > > near the dpi recommended for materials that need to be OCRed. The > quality of > > images from cameras would be not only low, but the OCR (even with the best > > software) would probably have many errors. For someone scanning items at > > home this might be ok, but for archival quality, I would not recommend > > cameras. If you are grant funded and the grant provider requires a certain > > level of quality, you need to make sure the scanning mechanism you use can > > scan at that quality. > >I know very little about digital cameras, so I hope I get this right. > >According to Wikipedia, Google uses (or used) an 11MP camera (Elphel >323). You can get a 12MP camera for about $200. > >With a 12MP camera you should easily be able to get 300 DPI images of >book pages and letter size archival documents. For a $100 camera you >can get more or less 300 DPI images of book pages. * > >The problems I have always seen with OCR had much to do with alignment >and artifacts than with DPI. 300 DPI is fine for OCR as far as my >(limited) experience goes - as long as you have quality images. > >If your intention is to scan items for preservation, then, yes, you >want higher quality - but I can’t imagine any setup for archival >quality costing anywhere near $1000. If you just want to make scans & >full text OCR available, these setups seem worth looking at - >especially if the software & workflow can be improved. > >best, >Erik > >* 12 MP seems to equal 4256 x 2848 pixels. To take a ‘scan’ (photo) of >a page at 300 DPI, that page would need to be 14.18" x 9.49" (dividing >pixels / 300). As long as you can get the camera close enough to the >image to not waste much space you will be getting in the close to 300 >DPI range for images of size 8.5" x 11" or less. >;; Erik Hetzner, California Digital Library >;; gnupg key id: 1024D/01DB07E3