With the increasing number photos captured day-in and day-out and with their ever improving qualities, Compression is, was and will always be a burning issue to be addressed.
Researchers at Google seem to have taken an inspiration from the HBO Comedy Silicon Valley’s Pied Piper for its latest developments in Compression. The group has developed a way to use neural networks, that work like a human brain to compress images better than the existing traditional methods.
Built using the tech giant’s open source Machine Learning library, TensorFlow, the team at Google has trained an AI system to learn how compression works by with the help of about 6 million photos, broken into 32 x 32 pixel pieces and then selecting 100 pieces from each of the images with the least effective compression to learn from. The crux of using the difficult parts for training, as researchers theorize, is to make the neural nets more prepared to handle the easy patches.
The AI system itself then predicts how the image would like like after it would be compressed and then generates that image. Previous research on the same topic that was published earlier by Google, proved to useful for images limited to a size of 64 x 64 pixels. This system is not limited by the size of the file.
The system is nowhere near to Pied Piper’s compression, which makes the file so small that its size is negligible, but the efforts made by Google promise that the unachievable can be achieved.
Its just the beginning, the system needs to be made better, to make it perfect in compressing images. But the work done by the team is really promising and it would be great to see them achieve more through the project in times to come.