An End-to-End Administrative Document Analysis System

Abstract

This paper presents an end-to-end administrative document analysis system. This system uses case-based reasoning in order to process documents from known and unknown classes. For each document, the system retrieves the nearest processing experience in order to analyze and interpret the current document. When a complete analysis is done, this document needs to be added to the document database. This requires an incremental learning process in order to take into account every new information, without losing the previous learnt ones. For this purpose, we proposed an improved version of an already existing neural network called Incremental Growing Neural Gas. Applied on documents learning and classification, this neural network reaches a recognition rate of 97.63%.

Tsallis Mutual Information for Document Classification

“…Nowadays, this task is especially important because huge volumes of documents are scanned to be processed in an automatic way. Some automatic solutions based on optical character recognition (OCR), bank check reader, postal address reader and signature verifier, have already been proposed but a lot of work has still to be done to classify other types of documents such as tabular forms, invoices, bills, and receipts [20]. Chen and Blostein [21] presented an excellent survey on document image classification.…”

Semantic information extraction from images of complex documents

“…Alternatively, we have considered lines and logos as noise to be removed, and focused our document image analysis on text elements, which has proven to be more robust and computationally efficient. -In other works [1,11,12,16] we find the utilization of Attributed Relational Graphs to represent document models, assuming that connecting nodes exactly above, below, left or right of each other can be detected. We have observed that such regularity is not observed in practice in many situations.…”

“…Hamza et al [11,12] have built a system, further developed by Belaïd et al [1], in which features are extracted and grouped together using a bottom-up approach. Keywords are segmented and grouped together with neighbouring keywords generating a high level structure.…”

“…As a consequence, different instances of a same document type can generate very different Attributed Relational Graphs, and the regularity of Attributed Relational Graphs to match a document model with document instances cannot be assumed a priori. -In many of these works [1,9,11,12,16,18] approximate solutions to known hard problems (such as graph distance, tree edit distance and maximum cliques) have been used to match an input document with a document model. These approximate solutions have proven to be sufficiently precise and sufficiently tractable to be used in practical situations.…”

Paving the way for next generation data-stream clustering: towards a unique and statistically valid cluster structure at any time step

“…We are thus faced with a 'plasticity/stability' dilemma -can a GNG network learn new data without using the data it learnt before and without deterioration of the network that has already been generated? An incremental GNG was developed (Prudent and Ennaji, 2004) and improved by Hamza et al (2008). It allows data to be learnt in parts if the database is too large to be learnt in one go and thus allows the learning process to be restarted even if the data learnt previously is no longer accessible.…”