An efficient parametrization of character degradation model for semi-synthetic image generation

Abstract

This paper presents an efficient parametrization method for generating synthetic noise on document images. By specifying the desired categories and amount of noise, the method is able to generate synthetic document images with most of degradations observed in real document images (ink splotches, white specks or streaks). Thanks to the ability of simulating different amount and kind of noise, it is possible to evaluate the robustness of many document image analysis methods. It also permits to generate data for algorithms that employ a learning process. The degradation model presented in [7] needs eight parameters for generating randomly noise regions. We propose here an extension of this model which aims to set automatically the eight parameters to generate precisely what a user wants (amount and category of noise). Our proposition consists of three steps. First, Nsp seed-points (i.e. centres of noise regions) are selected by an adaptive procedure. Then, these seed-points are classified into three categories of noise by using a heuristic rule. Finally, each size of noise region is set using a random process in order to generate degradations as realistic as possible.

Automatic Selection of Parameters for Document Image Enhancement Using Image Quality Assessment

“…The dataset was increased to 500 images by adding noise using the degradation model discussed in [18]. Some sample images are shown in Figure 2(a) & (b).…”

“…2) Hindi Dataset: The Hindi dataset contains a total of 470 text paragraphs extracted from grayscale document images [21]. In order to introduce variations in terms of noise and degradation, we add noise as cuts and merge to images sing the degradation model discussed in [18]. Some sample images are shown in Figure 2(c) & (d).…”

A Method to Generate Synthetically Warped Document Image

“…Such methods can roughly be classified into three types [12]. They are (i) adding noise, [10] (ii) degrading characters, [13] and (iii) distorting the shape of document images [12]. Kieu et.…”

Neural Networks for Document Image and Text Processing

“…Using synthetic data or synthetically degraded data has many advantages over human supervision including rapid generation of datasets at lower cost, control of degradation level, and fit testing of the same underlying document content with different corruption methods [Baird 2007;Kieu, Visani, Journet, Mullot, et al 2013;Varga et al 2003;Zi et al 2004]. The main idea is to take a clean image as the ground truth and apply several distortions and noise on top of it.…”

“…In particular, DIB evaluation is usually computed at pixel level (Section 4.4), and it requires an accurate ground truth, with the inherent complexity of data supervision at this detail level. Generating synthetic data for training and evaluating document image processing systems is a topic that has been widely addressed in recent years [Baird 2007;Kieu, Visani, Journet, Mullot, et al 2013;Varga et al 2003;Zi et al 2004]. To overcome this issue, there are several techniques to generate useful ground truths.…”