Method takes on determination of optimal image compression ratio

The emergence of teleradiology has made lossy compression techniques necessary to reduce image volume and accelerate communication - often at the expense of image quality.

Task-dependent and time-consuming visual thresholds or acceptance levels for lossy compression have evolved (normalized root mean square error or mean square error). But these have been found not to correlate well, and they do not provide visual quality information, according to Keh-Shih Chuang, Ph.D., of the nuclear sciences department at National Tsing-Hua University in Taiwan.

"This is due to the fact that human eyes extract structural information from the viewing field," said Tzong-Jer Chen from the same department.

Chen, Chuang, and colleagues have developed a quantitative image quality measurement using Moran statistics.

It has been shown (Chen et al. Phys Med Biol 2003;48(8):N131-137) that Moran statistics can be used as a quality index for measuring sharpness or smoothness of an image, since most image processing techniques alter the smoothness of the image. In a new paper (J Digit Imaging. 2003 Oct 2 [Epub ahead of print]), Moran statistics are used to quantify the blurriness and sharpness in response to compression ratio.

Using this method, a quality degradation model can be proposed for various image modalities, Chuang said.

"Our results suggest that one can determine the optimum ratio for image compression for teleradiology or image archiving," he said. "The image can be compressed at maximum ratio without degrading image quality."

In this technique, the amount of quality degradation as a function of compression ratio can be formulated as the sum of two terms representing edge blurring and denoising effects.

Blurring is related to structural complexity or contrast in the image content, according to Chuang. Its value increases with increasing compression ratio.

The effect of denoising on image quality is described by a negative exponential term, whose value is determined by image noise content. Denoising decreases as compression ratio increases and becomes negligible at a certain ratio, he said.