Form similarity via Levenshtein distance between ortho-filtered logarithmic ruling-gap ratios

Abstract

Geometric invariants are combined with edit distance to compare the ruling configuration of noisy filled-out forms. It is shown that gap-ratios used as features capture most of the ruling information of even low-resolution and poorly scanned form images, and that the edit distance is tolerant of missed and spurious rulings. No preprocessing is required and the potentially time-consuming string operations are performed on a sparse representation of the detected rulings. Based on edit distance, 158 Arabic forms are classified into 15 groups with 89% accuracy. Since the method was developed for an application that precludes public dissemination of the data, it is illustrated on public-domain death certificates.

CONFIRM – Clustering of noisy form images using robust matching

“…Form images have been represented in a large variety of ways for classification tasks (see [5] for a survey). These representations include statistics of image connected components [37], BoVW [6,24,38], OCR features [2,32,33], pyramids of average gray-scale values [18,32], Viola-Jones features [34], Hidden Tree Markov Models [10], sequences of line segments [12,20], sequence of line gap ratios [28], run length histograms [16], Shape Context Features [22], and most recently learned features from Convolutional Neural Networks [17,21,38].…”

“…In [12,20,28], forms are represented as sequences of vertical and horizontal rule lines, which are compared using a similarity metric such as edit distance or clique finding in an association graph. While these methods discretize or ignore the position or length of lines, CONFIRM performs a novel edit distance directly on a continuous representation of line segments, making it more robust to line detection errors.…”

“…While form recognition is a well-studied problem (e.g. [5,10,16,20,28,32,37]), fewer works, have considered the task of form clustering. The goal in this case is to partition the collection by form types (not known apriori ), which we define to be the exact layout structure of the form encompassing preprinted lines, text, and figures.…”

On Parallel Lines in Noisy Forms

“…Preliminary results on classification of some degraded forms were presented at the 2014 SPIE Conference on Document Recognition and Retrieval [38].…”