This article has been written to give a brief overview of the possibilities offered by Dox Premium. We have conduseveralal tests on a standard invoice to verify its limitations. This is the basic model we'll be using for our tests:
Experiment 1: Optimal Conditions
As depicted in the following images, nearly all fields ary recognized except for the mobile number. However, it's worth noting that all the identified information is accurate.
| Picture taken | Outcome |
| 
|
With good exposure and a clear image, Dox can easily recognize most fields. The only unrecognized field is the phone number, which is not labeled, and with a fictitious identifier, we can hope it will be better when it's more identifiable.
Experiment 2: Wrinkled Paper
In this scenario, we successfully identified all fields with the correct data. Similar to the previous attempt, however, the phone number remained elusive.
| Picture taken | Outcome |
| 
|
With good exposure and a clear image, Dox can easily recognize most fields. The only unrecognized field is the phone number, which is not labeled, and with a fictitious identifier, we can hope it will be better when it's more identifiable.
Experiment 3: Wrinkled paper + handwriting
Dox successfully recognized my handwriting once, but in other instances, it struggled to identify both the location of the text and its content. While the software efficiently handles printed text, it demonstrates limitations when dealing with handwritten input.
| Picture taken | Outcome |
| 
|
The results are quite good, with a clear recognition ofthe firstiting for the first one, but when the original field is not totally erased, Dox is hesitating between the fields it should focus on and returns a wrong number.
Experiment 4: Coffee stain
Due to the new challenge, not all fields are being recognized. Interestingly, those within the coffee stain are accurately identified, yet the issue lies with the area surrounding the stain. Dox prioritizes processing the challenging section but overlooks crucial fields such as the buyer's name, supplier's address, or even the account name.
| Picture taken | Outcome |
| 
|
Field recognition in the coffee stain is rather surprising, in terms of accuracy and field recognition. However, we can note a lesser performance concerning recognition outside the task. When faced with a significant difficulty, Dox focuses more on the difficulty than on its surroundings, and some information needs to be better processed.
Experiment 5: Torn sheet
It doesn't seem to disrupt the document processing in areas of difficulty. However, regarding the coffee stain, errors occur precisely around the challenging spots. In this instance, the software fails to detect the buyer's address, phone number, supplier's name, and account number.
| Picture taken | Outcome |
| 
|
We can draw the same conclusions as with the coffee stain. Data processing is still satisfactory.
Experiment 6: From far
In this latest attempt, nearly all fields are recognized successfully. However, some are missing, such as the subtotal. Additionally, if the printings are too small, they may not be recognized by the software.
| Picture taken | Outcome |
| 
|
Nothing unexpected here; some fields are barely readable. Nonetheless, Dox has still managed to find some of the information we were seeking.
Conclusion:
In conclusion, Dox Premium proves highly effective under optimal conditions. However, it encounters challenges in low-light settings or when dealing with small fields. To maximize Dox's potential, capturing a photo close to the object with proper exposure is recommended. While damaged documents can still be processed, the accuracy of results may vary across fields or might not be recognized at all.
- SAP Managed Tags:
1 Comment
