Deep-CNN Architecture for Error Minimisation in Video Scaling

摘要

People like to watch high-quality videos, so high-resolution videos are in more demand from few years. The techniques like DWT which are used to obtain the high-quality videos results with high distortions in videos which ends in low resolution. But there are numerous super-resolution techniques in signal processing to obtain the high-resolution frames from the multiple low resolution frames without using any external hardware. Super-resolution techniques offer very cheap and efficient ways to obtain high-resolution frames. Convolutional Neural Networks (CNN) technique is most widely used Deep-Learning technique for various application such as feature extraction, face detection, image classification, image scaling etc. Removing a noise from the frame is a very difficult task, so a CNN is introduced with super-resolution technique. Moreover, CNN technique can easily train bulky datasets, remove blurriness and can provide the end-to-end mapping between high and low-resolution patches. Therefore, here, we have introduced an efficient and robust Reconstruction Error Minimization Convolution Neural Network Architecture. Here, our model is highly efficient to handle large datasets and provide visually attractive results compared to existing state techniques using CNN architecture. The proposed CNN model has an additional unit of Pipelined structure to increase the processing speed operating on large datasets. Our experimental results verify that our model outperforms other existing state of-art-techniques in terms of Peak Signal to Noise Ratio-PSNR, Structural Similarity Index Matrices - SSIM and visual quality appearance.