The NoisyOffice Database: A Corpus To Train Supervised Machine Learning Filters For Image Processing

Abstract

This paper presents the ‘NoisyOffice’ database. It consists of images of printed text documents with noise mainly caused by uncleanliness from a generic office, such as coffee stains and footprints on documents or folded and wrinkled sheets with degraded printed text. This corpus is intended to train and evaluate supervised learning methods for cleaning, binarization and enhancement of noisy images of grayscale text documents. As an example, several experiments of image enhancement and binarization are presented by using deep learning techniques. Also, double-resolution images are also provided for testing super-resolution methods. The corpus is freely available at UCI Machine Learning Repository. Finally, a challenge organized by Kaggle Inc. to denoise images, using the database, is described in order to show its suitability for benchmarking of image processing systems.

A Block Permutational Steganographic Algorithm for Scanned Documents and other Images

“…We measure the detection rates for our algorithm. Results are shown in Table 2 and Table 3, for the NoisyOffice dataset of scanned images [36][37][38] and the MSRCv2 dataset of more generic images [39], respectively. The main measurement of detectability is the Area under the Receiver Operating Curve (AROC).…”

“…Table 1 lists the average computation times, both of our algorithm and of [35], for the MSRCv2 [39] and NoisyOffice [36][37][38] datasets. Tests were run on an Intel 7th Generation Core i7 notebook computer.…”

“…The steganography methods are trained and tested on two datasets: the NoisyOffice [36][37][38] dataset of scanned documents, and the MSRCv2 [39] dataset of more general images. The two datasets are evaluated separately.…”

“…In addition to the Wavelet Motion Analyzer, we also use additional methods to compare the algorithms: the Sample Pair Analysis [25] and Peak Signal Noise Ratio (PSNR) [3]. For both the NoisyOffice dataset [36][37][38] of scanned documents, and the MSRCv2 dataset [39] of more general images, Wavelet Motion Analyzer results indicate that the algorithm of [35] is outperformed by our own, while the other measures are either roughly comparable or improved.…”

“…Fig.5. Area Under the Receiver Operating Curve (AROC) results graphed against bit-per-pixel rate for the NoisyOffice dataset[36][37][38]. The lower the AROC, the less detectable the steganography method.…”