Performance Comparison of Six Algorithms for Page Segmentation

This paper presents a quantitative comparison of six algo- rithms for page segmentation: X-Y cut, smearing, whitespace analy- sis, constrained text-line finding, Docstrum, and Voronoi-diagram-based. The evaluation is performed using a subset of the UW-III collection commonly used for evaluation, with a separate training set for parame- ter optimization. We compare the results using both default parameters and optimized parameters. In the course of the evaluation, the strengths and weaknesses of each algorithm are analyzed, and it

is shown that no single algorithm outperforms all other algorithms. However, we observe that the three best-performing algorithms are those based on constrained text-line finding, Docstrum, and the Voronoi-diagram. 1 Introduction Document image layout analysis is a crucial step in many applications related to document images, like text extraction using optical character recognition (OCR), reflowing documents, and layout-based document retrieval. Layout analysis is the process of identifying layout structures by analyzing page images. Layout struc- tures can be physical (text, graphics, pictures, ...) or logical (titles, paragraphs, captions, headings, ...). The identification of physical layout structures is called physical or geometric layout analysis, while assigning different logical roles to the detected regions is termed as logical layout analysis [1]. In this paper we are concerned with geometric layout analysis. The task of a geometric layout anal- ysis system is to segment the document image into homogeneous zones, each consisting of only one physical layout structure, and to identify their spatial relationship (e.g. reading order). Therefore, the performance of layout analysis methods depends heavily on the page segmentation algorithm used. Over the last two decades, several page segmentation algorithms have been proposed in the literature (for a literature survey, please refer to [1,2]). The problem of automatic evaluation of page segmentation algorithms is in- creasingly becoming an important issue. Major problems arise due to the lack of a common dataset, a wide diversity of objectives, a lack of meaningful quantita- tive evaluation, and inconsistencies in the use of document models. This makes 2 the comparison of different page segmentation algorithms a difficult task. Mean- ingful and quantitative evaluation of page segmentation algorithms has received attention in the past. Yanikoglu et al. [3] presented a region-based page segmen- tation benchmarking environment, named Pink Panther. Liang et al. [4] proposed a performance metric for document structure extraction algorithms by finding the correspondences between detected entities and ground-truth. The quality of page segmentation algorithms was also evaluated by analyzing the errors in the recognized text [5]. However, text-based approaches have found little use since they measure the output of multiple steps and cannot be used to evaluate page segmentation alone. Das et al. [6] suggested an empirical measure of performance of a segmentation algorithm based on a graph-like model of the document. There has been little effort in the past to compare different algorithms on a quantitative basis. Mao et al. [7] presented an...

Website: reference.kfupm.edu.sa | Filesize: -
No of Page(s): 12
Download Performance Comparison of Six Algorithms for Page Segmentation.pdf