Semi-automatic Segmentation & Alignment of Handwritten

Performance with manually set parameters

The performance metrics for manually set parameters can be viewed in Table 3. It is seen that the IoU is greater for method 2 for both data sets. This indicates a better over- all performance for method 2 since the IoU value represents how close the segmentation is between an aligned word and that word in the GT, so both the segmentation and the alignment are accounted for. This is further fortified by the fact that the standard devia- tion is lower for method 2 IoU. By Looking at EsB1 and EsB2 it is clear that for Labour’s Memory, the number of boxes that method 1 segmented out is closer to the number of GT words since EsB1 is less than EsB2. Upon examining EaB1 and EaB2, it becomes ev- ident that method 2 exhibits a higher degree of alignment between the number of boxes and the GT, which ultimately holds greater significance. Labour’s Memory seems to be more prone to over-segmentation than IAM (Figure 16). The same thing can be said about method 2 compared to method 1 (Figure 16). The line segmentation seems to be equally as good between the two methods used.

Table 3: Performance metrics for manually set parameters. Data presented as mean (standard deviation). IoU=Intersection Over Union, 1=Method 1, 2=Method 2, EsB=Error segmentation boxes, EaB=Error alignment boxes, EW=Error words, EL=Error lines.

Data set IoU1 IoU2 EsB1 EsB2 EaB1 EaB2 EW1 EW2 EL1 EL2

Labour’s Memory 0.329
0.337
0.120
0.183
0.255
0.203
0.008
0.008
0.038
0.038

(0.038)(0.033)(0.112)(0.153)(0.108)(0.096)(0.037)(0.038)(0.067)(0.067)

IAM 0.745
0.814
0.176
0.105
0.181
0.126
0.094
0.044
0.006
0.006

(0.158)(0.078)(0.103)(0.052)(0.105)(0.063)(0.039)(0.023)(0.028)(0.028)




Figure 15: Distribution of IoU values for manually set parameters of the 20 tested images from both data sets for Method 1 and Method 2. a) Labour’s Memory, b) IAM.

Figure 16: Distribution of the fraction of the number of segmentation boxes and GT boxes for manually set parameters of the 20 tested images from both data sets for Method 1 and Method
2. The middle red line indicates a fraction of 1. a) Labour’s Memory, b) IAM.

2.