Analogic Preprocessing and Segmentation Algorithms for Offline Handwriting Recognition

Abstract

This report describes analogic algorithms used in the preprocessing and segmentation phase of offline handwriting recognition tasks. A segmentation-based handwriting recognition approach is discussed, i.e., the system attempts to segment the words into their constituent letters. In order to improve their speed, the utilized CNN algorithms, whenever possible, use dynamic, wave front propagation-based methods instead of relying on morphologic operators were embedded into iterative algorithms. The system first locates the handwritten lines in the page image, then corrects their skew as necessary. It then searches for the words within the lines and corrects the skew at the word level as well. A novel trigger wave-based word segmentation algorithm is presented, which operates on the skeletons of words. Sample results of experiments conducted on a database of 25 handwritten pages along with suggestions for future development are presented.

Novel Approach for Baseline Detection and Text Line Segmentation

“…A profile curve can be obtained by projecting black/white transitions or the number of connected components. The profile curve is then analyzed to find its maxima and minima [15,16,17]. In smearing technique, consecutive black pixels along the horizontal direction are smeared.…”

Cellular wave computer algorithms with spatial semantic embedding for handwritten text recognition

“…In the field of text image segmentation we strongly rely on the results published in [17] and partly in [18], which include pre-processing, line detection, word detection and upper and lower baseline extraction algorithms. Lines and words are detected by means of analogic algorithms based on the ones described in [17]. The process includes noise filtering, adaptive line detection, skew correction and word position detection ( Figure 2).…”

“…This section deals with wordlevel detection, line level is mentioned in the next one. Baseline extraction is performed by fitting lines on upper and lower extreme points of the word image by using analogic algorithms given in [17] (Figure 3). The method basically computes the pseudo-convex hull of the word using the HOLLOW template, and finds the lowest and uppermost points of the pseudo-convex hulls using the LOCAL SOUTHERN ELEMENT and the LOCAL NORTHERN ELEMENT (LNE) detector templates, respectively.…”

“…Baseline distance is measured between the upper and lower baselines (lines 2 and 3 in Figure 4) that are estimated using a method described in [17]. The algorithm is based on the same idea as the one described in Section 3 for the detection of baselines of words, but no refinement is made and it is applied to a whole line image.…”

“…Owing to our statistical approach the obtained width intervals are generally relatively wide, but in many cases incorrect lengths do not generate many word candidates, because they are excluded by illegitimate feature combinations when binding features to letter positions using shape codes (Section 8). However, combining the method described above with a segmentation technique that can validate letter segments (see algorithms in [17]), could lead to narrower interval estimates.…”