UCLA researchers in the Henry Samueli School of Engineering and Applied Science have created a new data compression technique that surpasses the capabilities of current techniques such as JPEG.
The team, led by Northrop-Grumman Optoelectronics chair professor of electrical engineering at UCLA, Bahram Jalali, created their technique through the realization that data could be compressed through stretching and warping the data by way of a mathematical function. One specific application for this data compressor was targeted toward “Big Data” in the science and medical field where there are massive amounts of data needing to be processed.
“Any digital object (piece of data) — a text file, a video, a picture; has a certain size, measured in bytes. Compressing that object allows you to represent it with a smaller amount of data. Compression is desirable for two main reasons. One: the compressed data takes up less space than the uncompressed data, so you can store more stuff in the same space. Two: because the compressed data has a smaller size, it is easier/faster to transmit over a network,” said John Owens, a professor in the Department of Electrical and Computer Engineering at UC Davis.
The drawback of many data compressors is the loss of data and detail once the object is compressed. Quality of images, for example, may be compromised for storage space or to fit in an email; through compression some of the details can be lost. However, with the UCLA technique, called “anamorphic stretch transform,” or AST, the quality of data is maintained.
“AST warps the images in such a way that it stretches the small features but not the large features. This warping causes more samples (bits) to be assigned to small features, where they are needed most, and less to large features where data would be redundant. So we end up with a smaller file size yet with high resolution to see the small object clearly. The beauty of it all is that it works without needing to know where the small features are. This important property is embedded in the physics of the process,” Jalali said.
The idea behind the AST comes from the artistic technique anamorphosis — a technique that involves the distortion of an image.
With compression speeds and clarity to surpass current techniques — such as JPEG — AST may be a solution to the problem of “big data.”
Mohammad Asghar is a postdoctoral researcher on the UCLA team.
“The large telescopes on earth capture 25 gigapixel images each second. Transmission and storing of this amount of information is just a headache. AST has a promising solution for such applications,” Asghar said.
Utilizing AST will open up a large realm of possibilities — increasing data compression allows for smaller files and sharper quality, thus many applications will benefit.
“It will make websites faster and will make it easier to email and share large files including medical images … This leads to more accurate image recognition, whether in finding a cancer tumor in a medical image or identifying a burglar in a security camera picture,” Jalali said.
Looking ahead for AST, the widespread usage of this data compressor is in the works, with the team continuing to fine-tune and deliver their creation.
“We need to optimize the algorithm to get the most compression with minimum amount of computation time. We also need to figure out how to make it available for everyone to use,” Jalali said.
