当前位置: X-MOL 学术IET Image Process. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Improved compound image segmentation using automatic pixel block classification with SVM
IET Image Processing ( IF 2.3 ) Pub Date : 2020-06-01 , DOI: 10.1049/iet-ipr.2018.6523
Ebenezer Juliet Selwyn 1 , Selvi Shunmuga Velayutham 2 , Jemi Florinabel Deva George 3
Affiliation  

Computer screen images such as wallpaper, web pages and powerpoint images are compound images. As these images contain a mixture of textual, graphical, pictorial and smooth regions, compression of computer screen images necessitates accurate classification and segmentation of these regions. In this study, an improved compound image segmentation using automatic block classification with support vector machine (SVM) is presented. First, the input compound image is divided into several non-overlapping blocks, and then the statistical features embedded in each block are mined after applying discrete wavelet transform. Then, the SVM model is trained effectively by using the informative samples of fuzzy c-means clustering, which takes block-level edge features and block neighbourhood information as input. Finally, the compound image is classified into two classes such as text/graphics and picture/background with the trained SVM model. Experimental results show that the proposed method performs automatic block classification with high accuracy. As the proposed classifier uses both structural and contextual information as features, block classification accuracy has been improved to a great extent. Hence, the proposed method has made ∼6.2% improvement in block classification accuracy while comparing with existing approaches.

中文翻译:

使用自动像素块分类和SVM改进的复合图像分割

诸如墙纸,网页和PowerPoint图片之类的计算机屏幕图像是复合图像。由于这些图像包含文本,图形,图片和平滑区域的混合,因此压缩计算机屏幕图像需要对这些区域进行准确的分类和分段。在这项研究中,提出了一种改进的使用支持向量机(SVM)的自动块分类的复合图像分割方法。首先,将输入的复合图像划分为几个非重叠的块,然后在应用离散小波变换后挖掘嵌入在每个块中的统计特征。然后,使用模糊c均值聚类的信息样本有效地训练SVM模型,该样本以块级边缘特征和块邻域信息为输入。最后,根据训练后的SVM模型,将复合图像分为两类,例如文本/图形和图片/背景。实验结果表明,该方法具有较高的准确率。由于提出的分类器同时使用结构信息和上下文信息作为特征,因此块分类的准确性得到了很大程度的提高。因此,与现有方法相比,该方法在块分类精度上提高了约6.2%。块分类的准确性已大大提高。因此,与现有方法相比,该方法在块分类精度上提高了约6.2%。块分类的准确性已大大提高。因此,与现有方法相比,该方法在块分类精度上提高了约6.2%。
更新日期:2020-06-01
down
wechat
bug