U. D. Dixit - Department of Electronics and Communication Engineering, BLDEA’s V. P. Dr .P. G. Halakatti College of Engineering & Technology, Vijayapura, India
M. S. Shirdhonkar - Department of Computer Science and Engineering, BLDEA’s V. P. Dr .P. G. Halakatti College of Engineering & Technology, Vijayapura, India

Recently most of the documents are authenticated by using a latent fingerprint impression. Examples of such documents are property registration, banking transactions, insurance documents, etc. The fingerprint-based document retrieval (FPDIR) has emerged to provide an easier way of accessing, browsing, or searching such document images. This paper proposes efficient fingerprint-based document image retrieval by employing multi-resolution Histogram of Oriented Gradient (HOG) features. The preprocessing technique presented in this paper employs a combination of top-hat and bottom-hat filtering operations to enhance the detected fingerprint image. Multi-resolution HOG features are constructed from horizontal, vertical and diagonal directional components of the enhanced fingerprint image. Finally, a standardized Euclidean distance metric is used as a tool for matching, ranking and retrieval of the document images. The proposed system is assessed by experimenting with a dataset of ۱۲۰۰ images. The precision and recall results obtained using the proposed research work have given an improvement of ۸% to ۱۴% in retrieval performance compared to earlier methods.

Document image retrieval, Top-hat filter, Bottom-hat filter, Histogram of Oriented Gradients, Discrete Wavelet transform