Hi
Is it possible to integrate OCR function in ODK at collect stage or aggregate, to extract text data such as respondent name, surname data directly from from the images (uploaded ID).
If yes, can you please guide how to implement the same.
Warm Regards
Ishwar Kumar
OCR accuracy is very hard to do, but it is possible. Let's move this thread to Features so it can be further discussed.
ODK Scan was an attempt to do exactly this on the mobile side.
Captricity is another solution that folks have used
Two questions to get the discussion started:
Can you tell us more about the high-level problem you are trying to solve or share a user-scenario or story about this feature?--
While going for surveys, surveyors instead of filling in respondent details just capture an photo of the ID, and in backend system shall be able to pick the details from the photo image and populate in sheets (excel, CSV etc.).
this will help reduce the time consumed on each survey and increase productivity and resource utilization.
Would something like ODK Scan or Captricity solve your problem? If not, what features are they lacking?
Both ODK scan and captricity are able to resolve the same in standalone, however believe there shall be option to integrate the same in ODK collect itself to have the data automatically extracted and saved in the forms.
this is basis on my understanding of current system, however there might be possibility that same is already possible in existing system, if so, will highly appreciate if you can guide me how to integrate the same.
Warm Regards
another usage of OCR probably: If the data is usually recorded on a book and I want to digitize it, I will still have hard copy and soft copy. In my country there are community based mother and children health monitoring. Usually they will conduct recording of weight, height, immunization. The record is a log book, but the data only available for that particular community. There is no further data processing since the log must be retyped to computer by higher level community and nobody will do that unless for critical usage. It is easier to digitize the data by using scanner or camera and convert it into text or tables.
I saw many outdated posts on ODK Scan and a single unanswered OCR Post.
Seeing as many of these posts seemed to trail off in 2015 and OCR has come a long way in the last decade, has there been any more recent developments on this or any renewed interest in the community to meet this need? Even if this wasn't incorporated into ODK Collect itself, I would think having a very simple external app like ODK Counter that could perform this function could be a game changer for some scenarios with lots of numerical manual data entry and improving quality control.
What high-level problem are you trying to solve?
There are a number of scenarios that I have where I am using an external device (usually a digital scale) for many repeat measurements. Most frequently it's taking mass. I would like to have illiterate/inumerate people be able to "scan" a screen more quickly, and more importantly, more accurately record/digitize these decimal values in a decimal/integer/text question without the need for manual entry.
Before being able to print barcodes I used a photo question with manually written numbers with a photo question and for the scenario above, I'm still just taking photos of the digital scale as a backup/verification source.
Any ideas on how ODK could help you solve it?
Without doing anything more involved within the photo/number questions, I think the easiest thing would be an external app that could feed strings into any necessary question.
In my free time, I've been trying to learn the basics of Flutter and this could be a fun project to develop an external Android/iOS app, but honestly have no idea if I'd be able to achieve this.
Upload any helpful links, sketches, and videos.
Here are some image examples of screens I'd be interested in getting into a integer/decimal question:
ChatGPT does a pretty good job of recognizing the numbers in that image. My guess is there will soon be an offline LLM that runs well on mobile that we can embed in Collect.
