Tech Stack
Links
Year
Status
The Challenge
Users needed an accurate and efficient way to input handwritten Kanji characters in the Japanese iOS app.
The Solution
Collaborated with another engineer on the research and development of a handwritten Kanji input feature. Led the investigation and selection of suitable CNN approaches by reviewing relevant academic papers and defining the model direction. Prepared and processed training data, including data collection, cleaning, and preprocessing, and supported the training of a CNN recognizing approximately 2,300 Kanji characters. The model, trained using Python, Keras, and TensorFlow, achieved 99.3% recognition accuracy. Contributed to implementation and supported integration of the trained model into the iOS app using Swift and Objective-C, enabling a high-accuracy and seamless handwritten input experience.
Interested in this stack?
Let's build something similar.