Our algorithm converts an image dataset
into tabular dataset
by extracting features of an image.
This algorithm is tested on simple problems like handwritten-digit recognition
, hand-sign recognition
etc and is able to extract features.
For complex problems like extracting features from a human face image, we have not tested it yet and it probably won't work due to complexity of the problem.
Output: -169.0
, -148.0
, -153.0
, -171.0
- Preprocess image to reduce computation
- Calculate transition matrix of the preprocessed Image
- Assign pixels to grids
- Assign reward to each state
- Find optimal point of the figure
- Calculate return
- Use it to extract features
- At the end we get a set of 4 features corresponding to each image
Inference Notebook link: https://www.kaggle.com/raj401/image-feature-extraction-using-rl
rewards | returns |
---|---|
Theory Notebook link: https://colab.research.google.com/drive/1ttKeYIf_TUjrm9HPu8U5_PRI2A7Vi6s9?usp=sharing