Fractal-Based Approach for Segmentation of Address Block in Postal Envelopes

Abstract

Abstract. In this paper, an address block segmentation approach based on fractal dimension F D is proposed. After computing the fractal dimension of each image pixel by the 2D variation procedure, a clustering technique based on K-means is used to label pixels into semantic objects. The evaluation of the efficiency is carried out from a total of 200 postal envelope images with no fixed position for the address block, postmark and stamp. A ground-truth strategy is used to achieve an objective comparison. Experiments showed significant and promising results. By using the 2D variation procedure for three ranges of neighbor window sizes (r = {3, 5}, r = {3, 5, 7}, and r = {3, 5, 7, 9}), the proposed approach reached a success rate over than 90% on average.

Measuring the Quality Evaluation for Image Segmentation

“…In order to test the accuracy of quality evaluation in real segmentation, the five discrepancy criteria were applied on two published approaches for postal envelopes; the first one based on feature selection in wavelet space [3] and the second one based on fractal dimension [4]. In both approaches, the same database composed of 200 complex postal envelope images, with no fixed position for the handwritten address blocks, postmarks and stamps was used.…”

“…According to [3], the wavelet-based segmentation rates are 97.36% for address block, 26.96% for stamps and 75.88% for postmarks. According to [4], the fractalbased approach rates are 97.24% for address block, 66.34% for stamps and 91.89% for postmarks.…”

“…By applying the five discrepancy criteria to [3] 's and [4] 's segmentation results, without separating the address block, stamp and postmark classes, we obtained the quality evaluation rates grouped in Table 1. This Table depicts that DBM CP − T ype I and DBSM SP and DBSM SP − II measures have the same sensibility than [3] 's and [4] 's address block evaluation. This fact means that these 3 measures were not able to accurately evaluate the results of real segmentation.…”