- A company needed to detect and recognize characters and numbers in the text of advertisements.
- To save server space, extracted information needed to be uploaded as text instead of as images.
- Detecting characters and numbers in different fonts and on a non-uniform background is particularly difficult.
- Developed a segmentation algorithm to localize text.
- Trained a machine learning model to detect text from images.
- Built an Android application to access a video stream from a server or capture an image, apply text detection and recognition, and upload text data to the server.
Image processing of live video
Identifying text from live video
Recognize characters and understand meaning
Store information on server
Generate analytics of the words in the advertisement
- Saved server space.
- Measured the effectiveness of advertisements on billboards and television.
- Saved time spent analyzing the advertisements manually.