|
Figure 7
Comparison of expert and automatic image classification trials for experiments L498 (a), LG36 (b), LN84 (c), LN83 (d) and LO19 (e). From the 2000-image source data, frames were divided into `Hit', `Maybe', `Miss' classes by a human expert, with class totals listed at the upper right of each panel. Separately, candidate Bragg spots were automatically segmented on each image by the DIALS spotfinder, thus producing a histogram of the number of Bragg candidates per image, with the number of spots plotted here on a logarithmic scale over the horizontal axis. A best-guess threshold for classifying images as hits or misses is indicated for each experiment as a vertical line, along with the number of images falling on each side of the threshold. The correspondence between expert and automatic classifications is indicated by blue, orange and yellow shading on the bar graphs. |