Autoencoder For Image Processing

Now a days autoencoders is popular in various field. But, there is twoapplications of autoencoders which is most popular: dimensionality reduction and information retrieval.

• By using these features we can perform various image processing.
• This project has four method of image processing
  • Generate Image
  • Noise Reduction
  • Constructing New Images
  • Colouring Images

Dataset:

• The dataset is from the below link: http://vis-www.cs.umass.edu/lfw/lfw.tgz

Code File:

• All the code is used inside the reecho_model.ipynb notebook to run image processing.
• The autoencoder model is saved as variational_autoencoder.py.

Report:

• The report on this project is saved as report.pdf.

Run Test:

• To run test you need to run the test_model.py python file.
• Dowload the project and run the below command in the same directory

---

python test_model.py

---

• Or, double click on the test_model.py file.

Results:

• Although the result is not promissing, the model works really well despite of lower amount of data and lower training time.
• By using some generated images and increase the image resolution we can achieve better result.
• AThe original and reconstructed data in 'image reconstruction' model
• The noise and reconstructed data in 'noise reduction' model
• The newly constructed data in 'generating new images' model
• The gray scale and reconstructed coloured images in 'colouring image' model