A hierarchical method for automated identification and segmentation of forms

Abstract

In this paper we propose a fully automatic hierarchical method for identification of forms using global as well as local features. Moments of certain orders are considered as global shape features and are utilised to reduce the search space by selecting a subset of forms present in the database. The type of the candidate form is then identified within this subset through detail analysis using local geometrical and topological features. The candidate form is then segmented to extract the user-filled information.

Identification of Very Similar Filled-in Forms with a Reject Option

“…This technique overcame drawbacks of filled-in forms, noise and distortions which are usual in faxed documents but it is not suitable to distinguish among very similar forms. Mandal and Chanda [4] proposed a hierarchical method for identification of forms that alleviates the problems coming from the handling of large sets of documents.…”

“…Hierarchical classifiers were proposed to tackle with large amount of document types with computational efficiency [3], [4] [7]. In more generic tasks and applications, hierarchical approaches including decomposition of features, as well as classifier combinations, are widely used in document classification nowadays [1].…”

Applications of Text Detection and its Challenges

“…[59] finds similarity by matching detected logos, lines and keywords. [50] uses a global shape feature based on 2nd and 4th order moments of projection profiles.…”

“…[51] gives an algorithm that can be used to extract tables from the form. In [50], form is segmented by finding the largest rectangular window that would fit over a detected line. In [16] all information other than filled in data is removed and words are segmented out based on separation, baseline, skew.…”

“…[6] uses a parallel thinning Heuristic block finding and tree construction [23] Window fitting over lines [50] Words segmented based on separation between connected components [16] Features for identification Lines [23] 2nd and 4th order moments of projection profiles [ Graph constructed using each frame as node [47] Matching Process for identification Difference between topologically sorted strings of input form graph and candidate form graph [47] Post Preprocessing…”