The SRCNN is a deep convolutional neural network that learns the end-to-end mapping of low-resolution to high-resolution images. Single image super resolution example has been tried to be created with Python-Keras and PyQt5. The goal of super-resolution (SR) is to recover a high-resolution image from a low-resolution input. For more technical details, you can check out this resource: https://appgenius.icu/abs/1501.00092)
SRCNN consists of these stages
- Preprocessing & Feature Extraction: High resolution images are converted into low resolution images. So, Up-scales LR image to desired HR size. Later extracts a set of feature maps from the up-scaled LR image.
- Non-Linear Mapping: Then it does mapping of LR and HR properties.
- Reconstruction: Produces the HR image from HR patches.
pip install keras
pip install opencv-python
pip install PyQt5
- opencv -> This is necessary because the SRCNN network was trained on the luminance (Y) channel in the YCrCb color space.
- pyqt5 -> For GUI only
- keras -> For SRCNN. SRCNN training will not be done, 3051crop_weight_200.h5 weight can be used instead. Also thanks to MarkPrecursor for training the model.
python main_GUI.py
Test with any image in the low_resolution_images directory
