JPEG is the best format to compress images. Well, not anymore! Google has now come up with a better format that is going to change digital images even more. But traditionally, this image system takes up much more space and is usually a relatively large file.
The essence of this new format is to make images clearer while reducing its size, thus making storage much easier. A bright, high-quality image often takes up a lot of space. Google has decided to use neural networks, which is a computer system modeled like the human brain and nervous system. It has turned out to be a great achievement as far as clearer images are concerned.
How it was developed
The Program named RAISR, uses artificial intelligence(AI) and is basically an up-scale system that works by enhancing the images without increasing the size of the image. But instead, it reduces their file size in the process.
During the development of the program, Google got hold of 6 million 1280×720 images randomly chosen images from the Internet. The images were then broken down into 32×32 tiles and then enhanced x100. This majorly involved the images with the worst compression ratios.
The main aim was to improve performance on the most sensitive data. Google knew that once this is achieved, it would be easier to succeed on any other compression front.
A learning process
Next, the researchers used the TensorFlow machine learning system that Google open-sourced last year to set the stage for experimental work on the neural network architectures.
The researchers then used around a million steps to train the program and collect a series of metrics. They then compared which set of training models produced the best compressing results.
How it works
If one has a low-resolution image, for example, RAISR upscales the image first with the use of a standard technique and then it applies a particular filter to each pixel of the image, depending on the surroundings and background of the image.
Just after its development, it appeared limited for use as a program, as it only change images of little resolution to one of a high resolution, but Google went ahead and improved it further.
For example, one can send a 1000×1500 image that takes up 100KB, and compress it to a 500×750 reduced quality image of 25KB. Additionally, one can apply RAISR to restore the sharpness of the image which has already been compressed
Still a work in progress
At the moment, the program only works on Google+, but plans are in high gear to introduce it to other apps and platforms shortly. This means that taking those pictures with a camera of poor quality should no longer be a worry, as the pictures can now be cleared using RAISR.
Another creative class is where the program can guess a missing part of a picture that is not clear or not there altogether. Using AI, the program figures it out and replaces the missing part of the picture in addition to enhancing the overall image quality.
