bit rate (bpp)

bit rate (bpp)

averages broken out by image type or view (left and right breast, CC and MLO view). This demonstrates the variability among various image types as well as the overall performance. Two sets of SNRs and bit rates are reported: ROI only and full image. For the ROI SNR the rates are identical and correspond to the nominal rate of the code used in the ROI. For the full images the rates vary since the ROI code is used in one portion of the image and a much lower rate code is used in the remaining background and the average depends on the size of the ROI, which varies among the images. A scatter plot of the ROI SNRs is presented in Fig. 6.

It should be emphasized that this is the SNR comparing the digital original with the lossy compressed versions.

5 Subjective Ratings

Subjective quality of a reconstructed image can be judged in many ways. A suitably randomized set of images can be presented to experts or typical users who rate them, often on a scale of 1 to 5. Subsequent statistical analysis can then highlight averages, variability, and other trends in the data. Such formalized subjective testing is common in speech and audio compression systems as in the Mean Opinion Score (MOS) and the descriptive rating called the diagnostic acceptability measure (DAM) [1,49,62]. There has been no standardization for rating still images.

A useful attribute of an objective quality measure such as SNR would be the ability to predict subjective quality. For medical images, it may be more important that a computable objective measure be able to predict diagnostic accuracy rather than subjective quality. A potential pitfall in relating objective

0 0

Post a comment