Lossless image compression
Article By : Andreas Breyer
Noise compression reduces RAW files down to a tenth of their size without loss of information.
The trend is clearly upward: Each new camera generation offers higher resolutions along with higher frame rates. The amount of data generated in the process creates a problem for scenarios in which the data must be stored over long periods of time. A Swiss startup applies its knowledge in quantum physics to the image sensor in order to reduce the size of RAW image files by the factor of 10 without any loss of information.
In scientific applications, the amount of image data can be large and the storage times long, which increasingly occupies IT systems and server workload of i.e., research institutions. Gerhard Holst, head of science & research at PCO (Kelheim, Germany), explained, “Big Data is a current problem which is heavily discussed in the area of scientific imaging. It is fueled by three significant developments: higher frame rate, higher resolution, and the requirement of creating three-dimensional recordings. Image data can generally be compressed, but previous approaches only allowed for compression after the image is already created, or with a low compression rate.” As a manufacturer of high-end camera systems researching new ways of compressing image data, PCO met Dotphoton, a Swiss startup specializing in image compression for critical applications and AI.
Dotphoton’s Jetraw software starts before the image is created and uses the information of the image sensor’s noise performance to efficiently compress the image data. The roots of the image compression date back to the research questions of quantum physics. For example, whether effects such as quantum entanglement can be made visible for the human eye.
Bruno Sanguinetti, CTO and co-founder of Dotphoton, explained, “Experimental setups with CCD/CMOS sensors for the quantification of the entropy and the relation between signal and noise showed that even with excellent sensors, the largest part of the entropy consists of noise. With a 16-bit sensor, we typically detected 9-bit entropy, which could be referred back solely to noise, and only 1 bit that came from the signal. It is a finding from our observations that good sensors virtually ‘zoom’ into the noise.”
Dotphoton showed that, with their compression method, image files are not affected by loss of information even with compression by a factor of ten. In concrete terms, Dotphoton uses information about the sensor’s own temporal and spatial noise. The specific measured data are therefore an input condition, which generated a welcome synergy effect, Holst said. PCO is a long-term supporter of the image processing standard EMVA 1288, which defines quality parameters of cameras so that they can be compared with each other. The measurement data required for the EMVA 1288 standard largely correspond to the required parameters in Dotphoton’s software and are already available in every PCO camera. “With Dotphoton’s solution, the information from the sensor and the camera can be used beneficially. It was extremely appealing from PCO’s point of view to obtain a compression method that exploits the individual image capture chain of each camera (model),” Holst explained. In the PCO system, the compressed image files are currently stored in the recorder module of the SDK and decompressed for processing by Dotphoton’s software. A fully integrated version of the software in the camera would be a logical next step, so that the camera only transfers the already compressed data. “The ultimate benefit would be an integration of the compression in the FPGA,” Holst said.
Dotphoton is convinced that this step is not too far away.
The first responses from PCO customers have been quite positive, other requests from applications such as Particle Image Velocimetry are currently under review. An expansion to other high speed imaging applications beyond the scientific sphere with equally high demands for data storage, such as crash testing, is also conceivable.
Another driving factor in Dotphoton’s development of Jetraw was the increasing usage of image data in AI applications. The number of images that were specifically generated for analytical purposes has increased drastically over the last few years, Sanguinetti said. “Many of these images are stored and serve the training of i.e., deep learning applications. This opens an opportunity for our compression algorithm because it aims to retain the metrological attributes of the RAW images. The information for camera calibration contained in Jetraw images are currently only used for compression. They could, however, also be used for the improvement of other image processing tasks, for example to enable machine learning to efficiently work with images from different sources.”
Dotphoton currently integrates various software packages and programming languages so that users can directly load and save Jetraw images. This already includes C/C++, Python, Java, and ImageJ. LabView and Matlab will be available soon as well.
This article was originally published on EE Times Europe.
