Single Classifier-Based Passive System for Source Printer Classification Using Local Texture Features

Abstract

Abstract-An important aspect of examining printed documents for potential forgeries and copyright infringement is the identification of source printer as it can be helpful for ascertaining the leak and detecting forged documents. This paper proposes a system for classification of source printer from scanned images of printed documents using all the printed letters simultaneously. This system uses local texture patterns based features and a single classifier for classifying all the printed letters. Letters are extracted from scanned images using connected component analysis followed by morphological filtering without the need of using an OCR. Each letter is sub-divided into a flat region and an edge region, and local tetra patterns are estimated separately for these two regions. A strategically constructed pooling technique is used to extract the final feature vectors. The proposed method has been tested on both a publicly available dataset of 10 printers and a new dataset of 18 printers scanned at a resolution of 600 dpi as well as 300 dpi printed in four different fonts. The results indicate shape independence property in the proposed method as using a single classifier it outperforms existing handcrafted feature-based methods and needs much smaller number of training pages by using all the printed letters.

Source Printer Classification Using Printer Specific Local Texture Descriptor

“…Elkasrawi is obtained by taking a majority voting on predicted labels corresponding to both 'e' and 'a'. In a more recent work, a single-classifier-based system was proposed in [10] which simultaneously uses information from all letters in a single classifier and does not need specific letters. This work uses connected component analysis followed by region separation and border removal.…”

“…The most comprehensive publicly available dataset [7] (DB1) and another existing dataset, created by us [10] (DB2), are used to evaluate the performance of different methods. DB1…”

“…The pooling parameter N p is chosen to be equal to the total number of occurrences of 'e' on a given printed document (as suggested in [10]). Under these settings, the proposed method (denoted as PSLTD e ) achieves an average page-level classification accuracy of 98.92% with a standard deviation of ±0.43 which is a gain of 1.8% over the best reported accuracy only on e's, using CC-RS-LTrP-PoEP e [10] (Table I). In addition, the state-of-the-art CNN based method (CNN-{S raw , S med , S avg } a,e [2]) gives an average page-level classification accuracy of 97.33% using all occurrences of 'e' and 'a' images along with their median and average residuals.…”

First steps toward CNN based source classification of document images shared over messaging app

“…So, first of all, the image is converted into a gray-scale image. All the letters (connected components) on a page are extracted using connected component analysis [15]. An extracted component C k is in the form of a tight box around a connected component.…”

“…An extracted component C k is in the form of a tight box around a connected component. Some spurious non-letter components such as punctuation marks are filtered out based on the size (height, width, and area) of a bounding box using the strategy proposed in [15]. We remove components having an area smaller than half of the median value of areas of all connected components on a test page.…”