求助:翻译下文,能翻多少是多少
8. Advance the top and bottom cutting lines down or up, respectively, provided that the difference is less than a threshold θ3 and the luminosity of the current cutting line does not significantly differ from that of its predecessor. (We are eating up a marginal artifact.)
9. Write out the portion of the image that is between the four cutting lines to a file.
4. Multiresolution segmentation method
Our method of segmenting and classifying regions of a page has been described previously in [15]. By using a multiresolution approach, we gain two benefits: (a) the essential image information for a structural analysis of the page is treated at a level of resolution in which high-frequency noise has been filtered out thereby preventing the introduction of small noise regions, and (b) speed is gained by performing most of the steps of analysis on reduced resolution representation of the image.
The algorithm proceeds in two phases. During the first phase, four feature pyramids are constructed from the document image. During the second phase, regions of the image are classified according to rules into categories: background, text, line-drawing, photograph, or unknown.
4.1. Phase 1: construction of feature pyramids
The first phase constructs four pyramids. First four “feature images” are computed from the original: a “mean” image, a median image, a variance image, and a “threshold” image. Let the width and height of the original image be m and n, respectively. The mean image of size m/16 by n/16 is obtained by computing the average pixel value for each 16×16 block of pixels in the original. The three other feature maps have the same dimensions as the mean image, but they contain the median, variance, and “threshold” values over the 16×16 blocks rather than the mean value. The threshold image is obtained by counting the number of pixels within each 16×16 block that exceed a threshold θ4=250 in the original image.
Each of the four feature pyramids is constructed from one of the feature images described above: one is based on average gray values; another is based upon variance; one is based upon median values; and the fourth is based upon the thresholding. The pyramids use a reduction factor of 4 in each of the horizontal and vertical directions and they are computed using Gaussian smoothing. In our experiments, the resulting pyramids contained 4 levels each.
4.2. Phase 2: classification of regions
Once the feature pyramids have been constructed, it is possible to classify pixels, and subsequently regions of the image, into categories corresponding to the principal kinds of document features in academic journal document images: background, text, line-drawing, and photograph. The classification makes use of fixed thresholds which we have chosen according to experiments with sample images.
First, we classify pixels as belonging to the background provided two criteria are met: the value of the pixel in the average pyramid is greater than θ5 and its value in the variance pyramid is less than θ6. We used θ5=240 and θ6=15. (These parameter values were chosen empirically—see [15].)
Second, we classify pixels as belonging to “graphics” (either line-drawings or photographs) by performing the following steps:
1. Segment the median image by computing the connected components of that image.
2. Determine the bounding box for each connected component having a value below the background value. These boxes we consider to be “regions”.
3. For each region, count the number of pixels in the threshold image whose value exceeds a threshold θ7. We denote this count by c. Let the corresponding area (in number of pixels) of the detailed image be denoted by a. We use θ7=170.
4. If the region is a graphics region, and if c/a 0.7 classify the region as line-drawing. Otherwise, classify it as photograph. (The thresholds θ5, θ6, θ7, θ8, θ9, θ10 and θ11 were chosen empirically in tests with over 100 document images). Additional details can be found in [15].
Third, we classify pixels as belonging to text regions by using the following steps.
1. Compute the connected components of the average image.
2. Determine the bounding box of each component. We consider these to be regions.
3. In each region, determine the number nt of “text pixels”. A text pixel must satisfy θ8 fave θ9 and θ10 fvar θ11. Here fave and fvar are the values of the pixel in the average and variance images, respectively.
4. Compute r=nt/a, where a is the total number of pixels in the region.
5. Classify each region as text if r 0.7, as uncertain if 0.3 r 0.7, and as unclassifiable if r 0.3.
5. Experimental results and discussion
Our methods were tested on all 125 of the images in the following portion of a now reasonably standard database of test images: the University of Washington English-Document Image Database I, Volume 2 “Grayscale Images”. This database contains gray-valued scans of English-language journal pages, and it has been used by other researchers (e.g. [14]) to test page segmentation methods.
The enhancement to eliminate print-through worked successfully on all images. Also, the marginal artifacts were successfully eliminated from all the images. In the case of partial-extra page elimination, all but one image were successfully processed; the one image whose partial extra page was not completely eliminated contained an unusual line drawing at the boundary of the page—most of the area of this unusual region was correctly eliminated but the program was not able to determine that all of this region should have been eliminated.
We illustrate the application of our techniques to three sample images from the set of images. In the first example, we show the processing of a typical page of text not containing figures. We show the original image, enhanced version in which print-through has been eliminated, and the version in which marginal artifacts and the partial extra page have also been removed. Next to each of these three versions are the results of applying the multiresolution segmentation procedure to them. In general, one can see that segmenting the original image without any treatment of the problems leads to a poor segmentation; in Fig. 1b, the entire page area has been classified as photograph. In Fig. 1d, the segmentation identifies two regions of text, but one of these regions corresponds to a partial extra page. In Fig. 1f, we can see that the region corresponding to the partial extra page has been eliminated.
Display Full Size version of this image (42K)
Fig. 1. (a) Original image of two columns of text with print-through, marginal artifacts and partial extra page (UW Document database image s041GRY), (b) segmentation of original, (c) enhancement of original in which print-through has been eliminated, (d) segmentation of enhanced image, (e) version of enhanced image having marginal artifacts and partial extra page eliminated, and (f) final segmentation.
Fig. 2 shows an example in which the original contains not only text but also a line drawing. Without cleaning, the segmentation produces a single “photo” region—an incorrect interpretation of the image. After eliminating print-through, the segmentation results in four regions (two textual, one photo, and one line drawing). Two of these are correct (the large text region and the line drawing). However, the photo region corresponds to a marginal artifact, and the text region on the right corresponds to a partial extra page. Fig. 2f shows the correct segmentation that results after the artifacts have been removed as shown in Fig. 2e.
[解决办法]
装个词霸漫漫看吧...
[解决办法]
装个词霸漫漫看吧...
[解决办法]
帮顶
[解决办法]
其实大家都很忙……
[解决办法]
我知道ＣＳＤＮ上的朋友是我最坚强的后盾！
请帮我看看:http://community.csdn.net/Expert/topic/5531/5531431.xml?temp=.6267969
[解决办法]
~-~
[解决办法]
分割页图像经文物影印及扫描油菜cinque , ,一个国会levialdia , 油菜lombardib和S tanimotoc一个dipartimento di scienze dell 'informazione大学罗马 " La Sapienza " ,途经salaria 113号 00198罗马,意大利二dipartimento di informatica e sistemistica , Pavia大学,经编辑1 , 27100帕维亚, 意大利的三部田绪. 和工程箱352350 ,西雅图华盛顿大学,佤族98195 ,美国收到1999年6月18日; 修改2001年2月16日; 接受2001年2月26日在网上2002年2月11日. 摘要分析扫描文件是非常重要的数字图书馆建设和无纸办公室. 一个重大的挑战是应付文物影印和扫描. 我们提出了一系列简单的技术处理这些困难. 用125图像华盛顿大学扫描文件数据库 我们展示的效果,这些方法在编写图像分割的多分辨率的算法. 作者关键词:文献分析; artifact消除; 分割; 打印通过; 边际效应; 局部额外页; 数字图书馆文章概要1 . 引言1.1 . 总动力1.2 . 问题描述2 . 以前的工作3 . 加工方法3.1 . 消除打印通过算法-1-[待遇打印通过] 3.2 . 边际文物和局部加页算法-2-[5治疗边际文物和局部加页] 4 . multiresolution分割方法4.1 . 第一阶段:建造金字塔特征4.2 . 第2阶段:区域划分5 . 实验结果与讨论5.1 . computational考虑6 . 结论提vitae 1 . 引言1.1 . 一般动机网上数字图书馆可以提供更好的信息分布和更灵活的访问途经搜索算法可以比 传统的印刷图书馆. 不过,加上现有的印刷材料,电子图书馆是一个昂贵,过程缓慢,除非好的自动化程序可以发展. 经过学术期刊的文章都复印和/或扫描,其约束,印刷版本, 各种器物常常被引进的图像作进一步分析. 不论这些文物需要拆除,然后再做进一步处理, 或特殊因素必须考虑以下的处理步骤,使他们能包容遗物. 我们解决了文物的开发手段来减少和/或消除他们从扫描图像文件之前 分割. 1.2 . 问题描述打印透过是因当印刷一侧的纸张是看得见的副本或 扫描对方. 它可造成版面分割算法虚假断定一个页包含一张照片时,它实际上包含 唯一文本. 边际文物造成的几个现象复印和扫描: ( 1 )曲率页离玻璃近约束力的出版物, ( 2 )影像边缘的页面背后,一个被扫, 由于歪斜,在当页的出版物是开放的, ( 3 )影像无效范围以外的新的一页, 或( 4 )影像封面的扫描仪或复印机超越边界的页. 另一个棘手的效应是存在着一个局部多页,当复印或扫描过程记录部分 页数面临的一个兴趣. 这些遗物通常会引起任何误解的地区,在随后进行分割或正确识别区域, 不属于该网页的利益,因此是无用的. 问题是发展的手段,消除不必要的文物这种方式,在以后的分割过程 正确地运作. 此外,这种方法应该验证一个大的和现实的数据库的文件图像. 本文组织如下:在第2 ,我们进行了调查,相关文献. 在第3条的分割,我们开发的描述. 在第4节我们介绍我们两个阶段的金字塔执行算法. 在第5我们报导实验结果,并提供了详细的讨论. 终于在第6我们给我们的结论. 2 . 以前工作的几个算法版面分割已提议于文献. 这些算法可以分为三类:自下而上,自上而下的方式和混合方式. tipical自下而上的算法是docstrum算法o 'gorman [1] ,游程平滑算法wahl et al . [2] Voronoi图算法的kise et al . [3] ,分割Jain和钰[4] 与文串分离算法fletcher和kasturi [5] 而自上而下的算法是X y切割算法斯蒂娜[ 6 7 ] 形状指示复盖算法的贝尔德[8]和贝尔德et al . [9]和算法的分类报纸块基于纹理分析王srihari〔10〕. 南风和周〔11〕建议混合使用一个分开合并策略. 一项调查可以发现o 'gorman和kasturi [12] ,唐et al . [13]和Jain和俞[4] . 自上而下的做法,开始与期望什么结构可能出现在首页, 并着手确定分子在相继finer层次的粒度. 在另一方面,自下而上的方法,通常是从个人像素或人物, 进而结合成较大的单位,如文字,线,图形元素等, 直到整页进行了分析. 成功,所有这些技术也是有限的,高质量的数字图像,输入到他们. 虽然法〔14〕许可直接手术unenhanced像素的扫描图像, 还患有轻微的文物和局部加页. 该方法目前我们自动清除文物影印及扫描可以与自己multiresolution 版面分割算法或任何其他系统.
[解决办法]
帮顶
[解决办法]
英文不太好
[解决办法]
上google翻译一下就可以啦，方便
[解决办法]
[解决办法]
这么多，怎么让你翻译啊
[解决办法]
