The Super Resolute (SR) model enhances the quality of a low-resolution image to a high resolution. The quality of the output image will be significantly improved and contain more details. The existing system uses deep neural network to enhance the image quality by applying the little filters many times in deep network which help get more contextual information over the low resolute image. The proposed model uses CNN (Convolutional Neural Network) to extract features from the input image and Residual patch data will be calculated. The proposed architecture uses shallow Neural Network which significantly enhances the computational time. It will extracts more information using residual patch map data to restore lost image details. The proposed model is better than the existing CNN models in terms of accuracy and computational time and it is significantly faster than existing models in terms of batch processing.
The proposed model uses BCDS300 dataset where dataset consists of 200 images for training model and 100 images for testing model. For the training process, the model uses mean square error (MSE) for error computation.
Second, in the system, the data loss is significantly lesser. The size of the feature map does not get reduced by the time when Convolutional operation is used, unlike the existing system. As in the model shallow neural network consisting of only 3 layers combined with input and the output layer due to which it reduces the data loss while training.
open training.ipynb file and run the code it will start the training process.
open output.ipynb file and change the inout file name and the output file name and the check the path correctly to run the cod. then it will create an output image wil the name provided in the path given