Repository navigation
Image Taking Guidelines
-
Review which process-mode you will be using. Pair-Photo processing has some restrictions, such as, photos must be taken in order: SuDoc then title page. Single-Photo processing works regardless of image ordering.
-
Make sure your device takes photos in an acceptable format. (JPEG, PNG, and many common photo formats)
iPhones may produce '.heic' photos, this format is not machine-processable. A file converter must be used before extraction, you can find one here.
- Read the image-taking guidelines, taking better photos results in more accurate records being acquired.
-
Images should also be taken in the order they should be processed in, as they will be ordered by the date created.
-
Note: There are no restrictions on the size of image files
-
Minimize extraneous text, all text in an image will be extracted.
-
Make sure that the final photos are properly oriented. Any photos that are not correctly rotated will be misinterpreted and result in no or bad results.
This is a particularly difficult document to photograph, it has a wide title and has non-SuDoc text below the Sudoc call number.
| Example Text | Image Example | Extracted Text |
|---|---|---|
| Misinterpreted Photo |  |
DOCS S 1.71/2 1981 supp. [printed guide] CU |
| Best Case Photo |  |
DOCS S 1.71/2 1981 |
- Half the text in the image is faded
- The background design may be misinterpreted as text
- Very blurry photo with bad lighting
- Some of the text in the back of the page adds noise to the image and distorts the extraction
- Very non-linear lines may throw off the ordering on the bounding boxes drawn around the text so while the words may be extracted correctly, they may not be in the correct order
- The extraction here should be fine but extra text such as the 'HUMBOLDT STATE COLLEGE' at the bottom will also be read and written
- The distortion on 'p95' makes the text difficult to read
- Text on a book spine cannot be read if it is vertical or not properly oriented in the photo
- This image could work if it were rotated so that the text was horizontal
- The text is cut off and the lighting adds further noise to the areas with text
- The contrast will distort the extraction on 'Statistical' and 'United'
- The spare text here will also be extracted
- Small text can be extracted but is much more error prone
