There have been selective communities in public which have wide usage of image enhancement systems, for personal use, commercial use, or academic use. In developing and operating such a system, there are five key parameters upon which the design must focus on: (1) up-scaling enhancement quality, (2) efficiency, (3) system acceptance, (4) operating platform, (5) ease of training and implementation.
Through the use of extensive image up-scaling and enhancement quality testing, system design analysis, and operating platform evaluation, our team has improved and optimized some critical aspects of image up-scale and enhancement system by adapting existing algorithm and developing our unique logistics. Hereby, we are presenting the communities with a low cost and easy adaptive system that is close to being ready for commercial development and high level applications.
Our filter design system is based on technology of machine learning method for image high/super resolution. By mapping the images with low and high resolutions from end to end, a convolutional neural network can be achieved. It has the ability to provide intelligent filtering technology to the program that performs the convolution to transform the low-resolution input into a high-resolution output. It will also employ image segmentation to aide the filter in edge detection This system can finish the process in the following steps:
- A process that takes a large set of images to train the model for the filter and output the filters
- Denoise the image
- Segment the image into partitions
- Scale each segment using bicubic upscaling by the desired scale value and run the filter
- Merge all upscaled segments into one final output image