I-Keyboard: Fully Imaginary Keyboard on Touch Devices Empowered by Deep Neural Decoder

We propose a fully imaginary keyboard (I-Keyboard) with a deep neural decoder (DND). Below are a few features of I-Keyboard.

The eyes-free ten-finger typing scenario of I-Keyboard does not necessitate both a calibration step and a pre-defined region for typing (first explored in this study!).
The invisibility of I-Keyboard maximizes the usability of mobile devices.
DND empowered by a deep neural architecture allows users to start typing from any position on the touch screens at any angle.
We collected the largest user data in the process of developing I-Keyboard and make the data public!
I-Keyboard showed 18.95% and 4.06% increases in typing speed (45.57 WPM) and accuracy (95.84%), respectively over the baseline.

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Requirements for Data Collection

MS Visual Studio >= 2015

Requirements for DND

CUDA >= 9.0
Python 3.6+
Tensorflow >= 1.14

Data Collection

Data Format

File Name: We have two types of file names.

D_A_S_I_T_P
- D: date (YYYYMMDD).
- A: age.
- S: sex (male or female).
- I: initial for discrimination.
- T: typing speed on physical keyboard.
- P: palm attached or detached while typing.
- e.g. 20190117_24_male_lhk_200_x
D_A_S_M_T
- M: major for discrimination.
- e.g. 20180831_26_male_enginerring_140

Data Format

One or two line(s) of phrases.
- When an enter is involved, two lines appear.
- The two phrases are separated by the enter key.
The sequence of x touch positions.
The sequence of y touch positions.

Data Visualization

To visualize the user behavior and analyze the statistics of user behavior, run the below. Running below will create two directories ('list_data' and 'figs') and save both the preprocessed results and the figures.

cd user_behavior_analysis

# preprocess the raw_data and save the result in the 'list_data' directory
python3 preprocessing.py

# visualize each experiment participant's typing behavior
# and extracts the statistics over the whole participants
python3 user_analysis.py

Training and Testing

For training, you need to set up a conda environment (recommended), or you can use pip instead

conda cread --name ikeyboard python=3.6
conda activate ikeyboard
conda install -c conda-forge tensorflow-gpu=1.14 editdistance

Then, you need to prepare data records using the "data.py" script as follows

python data.py

Executing the above script will generate a set of .tfrecords and .pkl objects required for training and testing.

Finally, you can train and test the propose DND model as follows

python train.py --name experiment_name
python test_experiment.py

For training options and test options, refer to "train_script.py" and "test_script.py".

Notification

The paper has been accepted! (IEEE Trans. on Cybernetics)
Any comments are welcome
Thank you for your attention

Citations

Please consider citing this project in your publications if you find this helpful. The following is the BibTeX.

@article{kim2019keyboard,
  title={I-Keyboard: Fully Imaginary Keyboard on Touch Devices Empowered by Deep Neural Decoder},
  author={Kim, Ue-Hwan and Yoo, Sahng-Min and Kim, Jong-Hwan},
  journal={IEEE Transactions on Cybernetics},
  year={2019}
}

Acknowledgement

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. NRF-2017R1A2A1A17069837).

Uehwan/I-Keyboard