"... In PAGE 4: ... The radius r is taken proportional to . The rules de ning the three segmentation class parameters are given in Table2 . These rules were used in the seg- mentation of a large variety images, ranging from sim- ple blood cells to complex indoor and outdoor scenes.... In PAGE 6: ... The analysis of the feature space is completely au- tonomous, due to the extensive use of image domain information. All the examples in this paper, and dozens more not shown here, were processed using the parameter values given in Table2 . Recently Zhu and Yuille [14] described a segmentation technique incorporating complex global optimization methods (snakes, minimum description length) with sensitive parameters and thresholds.... ..."

"... In PAGE 10: ... A more controversial issue is the optimal method for stem cell mobilization.141 As shown in Table3 the most common procedure includes the combination of high dose cyclophosphamide (HC-CTX) (2.5- 7g/m2) and G or GM-CSF.... ..."

"... In PAGE 3: ... The integer value modulo 3 will indicate whether to process a 5x5, 7x7 or 9x9 grid. The other integer argument speci es which one of the 9 parameters in Table1 to use in the computation. The function sdev computes the standard deviation of the grid of pixels surrounding the current pixel: rP (vi ? a)2 n where vi are the parameter values of the entries in the grid, a is the average of the grid area, and n is the number of entries in the grid.... In PAGE 3: ... The remaining oating point func- tions are the usual arithmetic functions. The oating point terminals include the twelve parameters from the image data in Table1 , and ephemeral random con- stants (Koza 1992). Integer expressions consist of either ephemeral random constants or the function inc, which increments an in- teger expression.... In PAGE 4: ... The tness value for a program is computed as: Fitness = 1 ? ce te cn tn where ce is the number of correctly identi ed edges, te is the total number of edges, cn is the number of cor- rectly identi ed non-edges, and tn is the total number of non-edges. 4 RESULTS Two 640 by 480 pixel thin section images of granitic gneisses were used, along with their ltered parame- ters of Table1 . Figures 1 and 2 shows their grey-scale maximum intensity cross polarized images (a), the in- tended target edge solution (b), and edge detection results (c, d, e).... ..."

"... In PAGE 19: ... Similarly,we consider values around each of the other two pairs of maxima. Table2 gives the results, which agree with those reported in Glasbey and Martin #281996#29, using an ad hoc similarity criterion. So, for example, we estimate that Fig 2#28b#29 needs to be shifted down by3 rows and shifted rightby 6 columns to align with Fig 2#28a#29.... ..."

"... In PAGE 8: ... These sets contain both 2-D and 3-D fluorescence microscope images taken from different cell types, as well as different micros- copy methods. Table3 summarizes the four image sets we used for the learning tasks described in this review. The 2-D CHO dataset was collected for five location pat- terns in Chinese hamster ovary cells.... ..."

"... In PAGE 16: ...2, while the practical results are obtained from Figures 3 and 4 respectively. Table3 provides an evaluation of the semantic segmenta tion results documented by the formation of the composite node within each of the figures. In the table, we present the normalized object size given by the ratio of pixels in the ground truth object to the total number pixels in the image, true positives (TP) normalized by the object size, and precision defined by TP/(TP+FP) where FP stands for false positives.... ..."