I had a breakthrough regarding Optical Character Recognition (OCR). In the first build of this computer vision app I'm planning to implement Google's on device ML Kit for document scanning. I've played with it before but I've wondered in the back of my mind if...
-
-
Prikaži ovu nit
-
I could use a similar approach to what I'm doing for object recognition for OCR. I mean, each character is just a single object. For my use case I only have to deal with one font in early versions, and a max of three total. The issue has been isolating the characters and today...
Prikaži ovu nit -
I came up with the broad brushstrokes of how to do that. I'll likely standardize my inputs, similar to normal NN's and such. This will be a bit of a pain, but will be very rewarding when it comes time to do algorithm development.
Prikaži ovu nit -
OCR will have to wait for awhile, but once I get to it I should have come up with a few more raycasting/computer vision tricks to deploy. It's pretty exciting because it could be an efficient enough approach to be worth generalizing further.
Prikaži ovu nit -
For instance, I could see getting to play with both differential graphics rendering and some topology stuff to handle loads of different fonts. These are things I haven't used in programming yet and it would be awesome to have a clearly defined problem to work on with them.
Prikaži ovu nit -
As I go into next week I'll get my head out of the clouds of solving future problems instead of right now problems. This week I'm going to decide if a C/C++ implementation is needed for Android, or if I can get away with something faster to deploy. As always, thanks for reading!
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.