There's nothing worse than having to search for a specific phrase in PDF files that have been "scanned" as images. But now Dropbox is coming to solve this problem by introducing a new automated image recognition tool to its cloud hosting service, which will automatically analyze the texts present in photos and PDFs, and add them as results to users' searches. According to Dropbox, there are currently more than 20 billion PDFs and photos.
To find the specific expression the user is looking for, all they have to do is enter it into Dropbox's file search engine, just like they would in any search engine. Dropbox will then return all PDF files that contain that specific word or phrase. The company said that this project is the most demanding undertaking machine learning department has attempted to implement to date.
The problems they encountered were quite a few. One of them is that PDF files with many pages were taking up a lot of computing resources of the system for a long time. So the team decided to set a setting where the index of each PDF is only performed for the first 10 pages of the file.
The automatic image recognition will only be able to recognize the English language, from JPEG, GIF, PNG, TIFF and PDF files that have been uploaded to Dropbox's cloud storage. The new service is expected to be released soon for enterprise packages, but it is not known when it will be implemented for ordinary users. Finally, the indexing process will also be performed on old files that have been uploaded to the service before the launch of the new feature.
