Whether it is a local or network hard disk cloud, we want to store the maximum amount of data with a minimum of space. To this end, a variety of file compression techniques and standards have emerged, there are pictures and jpeg png, mp3 audio there and flac, rmvb video there and mkv, as well as zip and tar and other compression . In fact, their ultimate goal is the same: to maximize their original information data, while generating a minimum volume of read-write file. Now, with Google neural network approach toward this goal and cross a big step forward.
A research team from Google recently with the neural network training method allows the compression efficiency of image files on the computer has reached a new height. According to foreign media reports, this new compression algorithm is better than traditional jpeg compression, in the same resolution, the use of neural network algorithms compress the file size of less than jpeg.
Specifically team is: First of 6 randomly selected from the Internet has been compressed over one million images, and each image is cut into a number of 32x32 pixel size pieces from each image are picked 100 compression the least effective debris input to the neural network learning. After repeated training, told the neural network which is the worst compression, let the computer learn the worst compression features. The scientists' idea is: If the computer master the worst characteristics of image compression, it can effectively avoid the subsequent compression algorithm, thus ensuring the file compression ratio at the same time, to ensure maximum image display.
The neural network compression to achieve a similar manner in the case of the training. Computer will be a photo firstly split into several small segments, and the characteristics of each segment individually compressed, and then subsequently combined to form the final output result of all segments. Team members publishedpaper, You can see the implementations of the algorithm, which also shows the neural network algorithm, in most compression ratio is better than jpeg effect.
But Google also acknowledged that the human visual system is perhaps the most difficult to grasp all aware of a system, because some people are very sensitive to certain changes, while others are not sensitive, is still unable to form a unified criteria, a picture in the end what makes the best display.
In any case, Google are trying our ever-growing media data as possible lossless compression to a minimum, which is a good thing at any time.