自监督学习的单幅透射图像恢复

现有基于学习的单幅透射图像恢复方法常需要大量成对的标签数据来训练模型,因缺乏成对图像集的监督约束,致使透射图像恢复效果欠佳,限制了其实用性.提出了一种基于自监督学习的单幅透射图像恢复方法,利用循环一致性生成对抗网络的循环结构和约束转移学习能力实现非成对图像的模型训练,通过设计自学习模块,从大规模的无监督数据中挖掘自身的监督信息对网络进行训练,以此形成有效的从浅层到深层的特征提取,提高透射图像正面内容的纹理、边缘等细节信息恢复质量,实现单幅图像的透射去除.实验结果表明,该方法在合成图像数据集、公共图像数据集以及真实图像数据集上都取得了较好的透射图像恢复结果.     

RSLDI: Restoration of single-sided low-quality document images

This paper addresses the problem of enhancing and restoring single-sided low-quality single-sided document images. Initially, a series of multi-level classifiers is introduced covering several levels, including the regional and content levels. These classifiers can then be integrated into any enhancement or restoration method to generalize or improve them. Based on these multi-level classifiers, we first propose a novel PDE-based method for the restoration of the degradations in single-sided document images. To reduce the local nature of PDE-based methods, we empower our method with two flow fields to play the role of regional classifiers and help in preserving meaningful pixels. Also, the new method further diffuses the background information by using a content classifier, which provides an efficient and accurate restoration of the degraded backgrounds. The performance of the method is tested on both real samples, from the Google Book Search dataset, UNESCO's Memory of the World Programme, and the Juma Al Majid (Dubai) datasets, and synthesized samples provided by our degradation model. The results are promising. The method-independent nature of the classifiers is illustrated by modifying the ICA method to make it applicable to single-sided documents, and also by providing a Bayesian binarization model.     

A multi-scale framework for adaptive binarization of degraded document images

In this work, a multi-scale binarization framework is introduced, which can be used along with any adaptive threshold-based binarization method. This framework is able to improve the binarization results and to restore weak connections and strokes, especially in the case of degraded historical documents. This is achieved thanks to localized nature of the framework on the spatial domain. The framework requires several binarizations on different scales, which is addressed by introduction of fast grid-based models. This enables us to explore high scales which are usually unreachable to the traditional approaches. In order to expand our set of adaptive methods, an adaptive modification of Otsu's method, called AdOtsu, is introduced. In addition, in order to restore document images suffering from bleed-through degradation, we combine the framework with recursive adaptive methods. The framework shows promising performance in subjective and objective evaluations performed on available datasets.     

Enhancement of historical printed document images by combining Total Variation regularization and Non-local Means filtering

This paper proposes a novel method for document enhancement which combines two recent powerful noise-reduction steps. The first step is based on the Total Variation framework. It flattens background grey-levels and produces an intermediate image where background noise is considerably reduced. This image is used as a mask to produce an image with a cleaner background while keeping character details. The second step is applied to the cleaner image and consists of a filter based on Non-local Means: character edges are smoothed by searching for similar patch images in pixel neighborhoods. The document images to be enhanced are real historical printed documents from several periods which include several defects in their background and on character edges. These defects result from scanning, paper aging and bleed-through. The proposed method enhances document images by combining the Total Variation and the Non-local Means techniques in order to improve OCR recognition. The method is shown to be more powerful than when these techniques are used alone and than other enhancement methods.     

Document Ink Bleed-Through Removal with Two Hidden Markov Random Fields and a Single Observation Field

We present a new method for blind document bleed-through removal based on separate Markov Random Field (MRF) regularization for the recto and for the verso side, where separate priors are derived from the full graph. The segmentation algorithm is based on Bayesian Maximum a Posteriori (MAP) estimation. The advantages of this separate approach are the adaptation of the prior to the contents creation process (e.g., superimposing two handwritten pages), and the improvement of the estimation of the recto pixels through an estimation of the verso pixels covered by recto pixels; moreover, the formulation as a binary labeling problem with two hidden labels per pixels naturally leads to an efficient optimization method based on the minimum cut/maximum flow in a graph. The proposed method is evaluated on scanned document images from the 18th century, showing an improvement of character recognition results compared to other restoration methods.     

Automatic accurate broken character restoration for patrimonial documents

In this article, we are interested in the restoration of character shapes in antique document images. This particular class of documents generally present a lot of involuntary historical information that have to be taken into account to get quality digital libraries. Actually, many document processing methods of all sorts have already been proposed to cope with degraded character images, but those techniques often consist in replacing the degraded shapes by a corresponding prototype which is not satisfying for lots of specialists. For that, we decided to develop our own method for accurate character restoration, basing our study on generic image processing tools (namely: Gabor filtering and the active contours model) completed with some specific automatically extracted structural information. The principle of our method is to make an active contour recover the lost information using an external energy term based on the use of an automatically built and selected reference character image. Results are presented for real case examples taken from printed and handwritten documents.     

Adaptive inverse halftoning for scanned document images through multiresolution and multiscale analysis

This paper describes an efficient algorithm for inverse halftoning of scanned document images to resolve problems with interference patterns such as moiré and graininess when the images are displayed or printed out. The algorithm is suitable for software implementation and useful for high quality printing or display of scanned document images delivered via networks from unknown scanners. A multi-resolution approach is used to achieve practical processing speed under software implementation. Through data-driven, adaptive, multi-scale processing, the algorithm can cope with a variety of input devices and requires no information on the halftoning method or properties (such as coefficients in dither matrices, filter coefficients of error diffusion kernels, screen angles, or dot frequencies). Effectiveness of the new algorithm is demonstrated through real examples of scanned document images, as well as quantitative evaluations with synthetic data.     

文档图像基准生成系统

为生成含噪声的扫描文档图像的基准标引信息,系统首先基于无噪声的PDF文档抽取理想化标引信息,采用透视变换模型,将其与含噪声文档图像进行配准,最终生成含噪声图像的基准标引信息,将其用于测试文字识别、检索的精度.系统还基于几种经典的图像退化模型,批量产生了含不同噪声类型的文档图像.经实验表明,该系统标引信息精度高,图像退化结果与实际噪声效果接近.     

Automatic generation of structured hyperdocuments from document images

As sharing documents through the World Wide Web has been recently and constantly increasing, the need for creating hyperdocuments to make them accessible and retrievable via the internet, in formats such as HTML and SGML/XML, has also been rapidly rising. Nevertheless, only a few works have been done on the conversion of paper documents into hyperdocuments. Moreover, most of these studies have concentrated on the direct conversion of single-column document images that include only text and image objects. In this paper, we propose two methods for converting complex multi-column document images into HTML documents, and a method for generating a structured table of contents page based on the logical structure analysis of the document image. Experiments with various kinds of multi-column document images show that, by using the proposed methods, their corresponding HTML documents can be generated in the same visual layout as that of the document images, and their structured table of contents page can be also produced with the hierarchically ordered section titles hyperlinked to the contents.     

Document Similarity Using a Phrase Indexing Graph Model

Document clustering techniques mostly rely on single term analysis of text, such as the vector space model. To better capture the structure of documents, the underlying data model should be able to represent the phrases in the document as well as single terms. We present a novel data model, the Document Index Graph, which indexes Web documents based on phrases rather than on single terms only. The semistructured Web documents help in identifying potential phrases that when matched with other documents indicate strong similarity between the documents. The Document Index Graph captures this information, and finding significant matching phrases between documents becomes easy and efficient with such model. The model is flexible in that it could revert to a compact representation of the vector space model if we choose not to index phrases. However, using phrase indexing yields more accurate document similarity calculations. The similarity between documents is based on both single term weights and matching phrase weights. The combined similarities are used with standard document clustering techniques to test their effect on the clustering quality. Experimental results show that our phrase-based similarity, combined with single-term similarity measures, gives a more accurate measure of document similarity and thus significantly enhances Web document clustering quality.     

一种改进的中文文档图像倾斜检测方法

图像获取设备将纸质文档转换为文档图像时，经常会使文档图像出现某种程度的倾斜，从而可能使后续的文档版面理解和OCR识别算法失败。文中提出一种基于近邻法的中文图像的倾斜角度检测方法，并采用最小二乘法减小倾斜估计的误差，从而大大优化了运算速度，增强了算法的鲁棒性，与现有方法相比，具有运算速度快，检测精度高的优势。算法在Visual C＋＋下编程加以实现，通过对检测库中100幅倾斜中文文档图像的检测证明，该方法具有精度高和适应性强的特点。     

Document image binarization based on topographic analysis using a water flow model

This paper proposes a local adaptive thresholding method based on a water flow model, in which an image surface is considered as a three-dimensional (3-D) terrain. To extract characters from backgrounds, we pour water onto the terrain surface. Water flows down to the lower regions of the terrain and fills valleys. Then, the thresholding process is applied to the amount of filled water for character extraction, in which the proposed thresholding method is applied to gray level document images consisting of characters and backgrounds. The proposed method based on a water flow model shows the property of locally adaptive thresholding. Computer simulation with synthetic and real document images shows that the proposed method yields effective adaptive thresholding results for binarization of document images.     

Document retrieval from compressed images

With the emergence of digital libraries, more and more documents are stored and transmitted through the Internet in the format of compressed images. It is of significant meaning to develop a system which is capable of retrieving documents from these compressed document images. Aiming at the popular compression standard-CCITT Group 4 which is widely used for compressing document images, we present an approach to retrieve the documents from CCITT Group 4 compressed document images in this paper. The black and white changing elements are extracted directly from the compressed document images to act as the feature pixels, and the connected components are detected simultaneously. Then the word boxes are bounded based on the merging of the connected components. Weighted Hausdorff distance is proposed to assign all of the word objects from both the query document and the document from database to corresponding classes by an unsupervised classifier, whereas the possible stop words are excluded. Document vectors are built by the occurrence frequency of the word object classes, and the pair-wise similarity of two document images is represented by the scalar product of the document vectors. Nine groups of articles pertaining to different domains are used to test the validity of the presented approach. Preliminary experimental results with the document images captured from students’ theses show that the proposed approach has achieved a promising performance.     

Mean-Shift segmentation and PDE-based nonlinear diffusion: toward a common variational framework for foreground/background document image segmentation

The presence of noise in images of degraded documents limits the direct application of segmentation approaches and can lead to the presence of a number of different artifacts in the final segmented image. A possible solution is the integration of a pre-filtering step which may improve the segmentation quality through the reduction of such noise. This study demonstrated that combining the Mean-Shift clustering algorithm and the tensor-driven diffusion process into a joint iterative framework produced promising results. For instance, this framework generates segmented images with reduced edge and background artifacts when compared to results obtained after applying each method separately. This improvement is explained by the mutual interaction of global and local information, introduced, respectively, by the Mean-Shift and the anisotropic diffusion. Another point of note is that the anisotropic diffusion process smoothed images while preserving edge continuities. The convergence of this framework was defined automatically under a stopping criterion not previously defined when the diffusion process was applied alone. To obtain a fast convergence, the common framework utilizes the speedup algorithm of the Fukunaga and Hostetler Mean-Shift formulation already proposed by Lebourgeois et al. (International Conference on Document Analysis and Recognition (ICDAR), pp 52–56, 2013). This new variant of the Mean-Shift algorithm produced similar results to the original one, but ran faster due to the application of the integral volume. The first application of this framework was document ink bleed-through removal where noise is stemmed from the interference of the verso side on the recto side, thus perturbing the legibility of the original text. Other categories of images could also be subjected to the proposed framework application.     

Unconstrained logo detection in document images

A fast and effective algorithm is developed for detecting logos in grayscale document images. The computational schemes involve segmentation, and the calculation of the spatial density of the defined foreground pixels. The detection does not require training and is unconstrained in the sense that the presence of a logo in a document image can be detected under scaling, rotation, translation, and noise. Several tests on different electronic document forms such as letters, faxes, and billing statements are carried out to illustrate the performance of the method.     

SOS：模糊图象反扩散恢复的“搜寻—单击”算法

为了对模糊图象进行高保真度的复原,研究了讨论了一种模糊图象的反扩散恢复算法;该算法首先以图象模糊的尺度为准,在比较为逐次递推算法和传统的单击算法的基础上,给出了它们的适应范围,进而提出了“搜寻－单击”算法,它用逐步逼近方式来探定未知的模糊尺度,然后以尺可能小的累积误差完成反扩散计算,从而实现了模糊图象恢复的盲处理,最后,采用自然模糊与人工模糊的图偈对恢复算法进行了验证,包括算法对模糊尺度的灵敏度算法的有效应用范围,以及纹理结构或景深偏差变化的模糊图象经恢复处理后的效果比较等,试验结果表明对数字图象作为SOS反扩散恢复处理可产生常规光学摄影技术所不可能取得的场景清晰度。     

Efficient document image binarization using heterogeneous computing and parameter tuning

In the context of historical document analysis, image binarization is a first important step, which separates foreground from background, despite common image degradations, such as faded ink, stains, or bleed-through. Fast binarization has great significance when analyzing vast archives of document images, since even small inefficiencies can quickly accumulate to years of wasted execution time. Therefore, efficient binarization is especially relevant to companies and government institutions, who want to analyze their large collections of document images. The main challenge with this is to speed up the execution performance without affecting the binarization performance. We modify a state-of-the-art binarization algorithm and achieve on average a 3.5 times faster execution performance by correctly mapping this algorithm to a heterogeneous platform, consisting of a CPU and a GPU. Our proposed parameter tuning algorithm additionally improves the execution time for parameter tuning by a factor of 1.7, compared to previous parameter tuning algorithms. We see that for the chosen algorithm, machine learning-based parameter tuning improves the execution performance more than heterogeneous computing, when comparing absolute execution times.

     

基于内容的文档图像倾斜校正

纸质文档通过图像获取设备转换为文档图像，由于人为因素和一些其它原因，文档图像不可避免地包含一定的倾斜角度。为了便于计算机处理，有必要对文档图像进行倾斜校正。文档版面十分复杂，包含文字、图像、图形、表格等内容。建立一个较为通用的文档图像倾斜校正算法是很困难的。提出了基于内容的文档倾斜自动校正方法，通过小波变换、游长平滑和细化处理，提取表格中的水平线和垂直线或文字行。针对不同的文档版面采用相应的倾斜校正策略。实验表明该方法具有倾斜校正速度快、精度高和适应性强的特点。     

Marginal noise removal of document images

Marginal noise is a common phenomenon in document analysis which results from the scanning of thick documents or skew documents. It usually appears in the front of a large and dark region around the margin of document images. Marginal noise might cover meaningful document objects, such as text, graphics and forms. The overlapping of marginal noise with meaningful objects makes it difficult to perform the task of segmentation and recognition of document objects. This paper proposes a novel approach to remove marginal noise. The proposed approach consists of two steps which are marginal noise detection and marginal noise deletion. Marginal noise detection will reduce an original document image into a smaller image, and then find marginal noise regions according to the shape length and location of the split blocks. After the detection of marginal noise regions, different removal methods are performed. A local thresholding method is proposed for the removal of marginal noise in gray-scale document images, whereas a region growing method is devised for binary document images. Experimenting with a wide variety of test samples reveals the feasibility and effectiveness of our proposed approach in removing marginal noises.