Document class recognition using a support vector machine approach

摘要

In most document archiving systems, one of the main fields is to identify the category of documents. In most case, determination of the document category in archiving tasks requires the application of classification model, which have had successes in improving documents processing. However, concerns exploding the frequency of use of documents in many office managers have driven increasing interests in Document Image Analysis (DAI) system. An automated tool that can distinguish photographs, textual and mixed documents in heterogeneous dataset can be effective to reduce the complexity measure in archiving process. Otherwise, instead of applying the same archiving strategy on the all documents dataset, a machine vision system can be used for identifying only the archiving process which can be used for each input document. This paper examines the use of support vector machine (SVM) algorithm for the helpful classification of documents in heterogeneous digital documents dataset. Our purpose is to investigate if a sufficient classification rate can be achieved when SVM is employed as a classification model in an automated document archiving system. In our experiments, a mixture of low-level features that characterizes documents in dataset was tested to build the aforementioned classification model. The obtained results reveal that the proposed SVM model yields 96% accuracy over a set of 250 test documents.