Inspiration
Our inspiration for this project was for translation applications. Copying a subtitle by hand is very time-consuming, and the existing solutions require large training datasets. We wanted to make a fast
What it does
Optical-Character-Recognition-For-HardCoded-Subtitles.io extracts pictures at a fixed sample rate from online video services. It detects subtitles present on the pictures, and extracts the words into a .srt file. It then corrects errors incurred in the transcription process, and provides an option to translate the results.
How I built it
We built the bulk of the project in python. We found the opencv, pytesseract, and google-cloud-translate libraries to be very useful in this project. To extract screenshots from video streaming services we used a chrome extension programmed in javascript.
Challenges I ran into
Designing the preprocessing for the optical character recognition was daunting, we tried nearly everything we could think of. It was almost unfortunate when our most successful solution used only 4 lines of code, as we had developed a number of interesting techniques.
Accomplishments that I'm proud of
Nobody had any experience with any of the packages we used, or had any experience at a hackathon or coding competition, so we feel pretty proud just of having a serviceable project.
What I learned
The Google cloud credits were a great way to explore different Google APIs. More or less everything was relatively novel to us, so we ended up learning a lot.
What's next for Optical_Character_Recognition_For_HardCoded_Subtitles.io
Our program still has some difficulty with non-monochromatic text, and text overlaid with artifacts. Logographic languages, such as Chinese, are also difficult. Working out ways to handle these obstacles would be the next step.
Built With
- google-cloud
- javascript
- python
- tesseractocr
Log in or sign up for Devpost to join the conversation.