As consumer products continue to grow quality wise and when the data usage of media is continually the demand for high resolution increases as well. This makes upscaling a very important part of the equation. Whether it is remastering old footage or upscaling from an input stream such as Netflix, upscaling has real world usage that is always more relevant than before.
Per databases, there has been a shear of interest in image up-scaling since 1993. It is because the establishment of World Wide Web in 1991, and the launch of T3 network standard in 1993, which enabled commercial users to transfer data at 44.736Mbit/s over 1.544Mbit/s via T1 standard. With the increase of accessible bandwidth and cheap storage options, people would not satisfy with the low-resolution footage on the internet, in addition, they would like to restore the information came in lost in details.
Although there are various up-scaling and enhancement algorithms that have already been developed, even being commercially sold, they are not perfect. A lot of them need to meet a performance requirement and often under perform as a result.
In order to solve this problem, our application aims to generate unique filter for particular assigned image category. This can be done by using Convolutional Neural Network. With the advantages of machine learning, we can easily process more detailed images and complex objects.
- Speed of machine learning
- Segmentation for edge detection and feature handling
- Cost and efficiency