Stamp Verification using Statistical Quantized Texture Histogram Features
Elnoordiansyah, Bambang Heru Iswanto, and Widyaningrum Indrasari

Department of Physics, Universitas Negeri Jakarta, Jl. Rawamangun Muka, Jakarta 13220, Indonesia

Abstract

The duty stamps used along with signature are the most widely used in official paper documents in Indonesia. However, non-genuine stamps are often found on the market and thus require techniques to identify their originality. In this paper, a simple method is proposed to verify the originality of the stamps using the statistical quantized texture histogram features and neural networks. The effectiveness of the method has been tested through a series of experiments using 75 seal samples, consisting of 25 original and 50 non-original stamps scanned with different resolutions. Euclidean distances between feature vectors in the feature space are analyzed using the principal component analysis (PCA) method. The analysis result showed that in the feature space, the separation of features from the genuine stamps and non-genuine stamps was very significant. Images with a resolution of 150 dpi have the highest feature dissimilarity with the smallest sum square error (SSE) compared to the others. The artificial neural networks trained with these instances succeeded in classifying the stamps with an accuracy and precision of around 90%. Therefore, the proposed method is accurate and robust to verify the stamp features

Keywords: statistical texture histogram, image recognition, component analysis

Topic: Computer Science