期刊名称：International Journal of Computer Science and Network Security
印刷版ISSN：1738-7906
出版年度：2018
卷号：18
期号：9
页码：79-84
出版社：International Journal of Computer Science and Network Security
摘要：CAPTCHAs are extensively used on the internet in order to distinguish between humans and bots. Numerous design alternatives of CAPTCHAs are introduced in last decade or so but still Text-based CAPTCHAs are most prevalent on the web due to its easy implementations. Text-based CAPTCHAs offer distorted text in images. The users are asked to read the text in order to prove them as humans. Although this is a trivial task for humans but it is still difficult for machines to decode such distorted characters with background clutter. The main challenges involved are preprocessing, segmentation and recognition. In this research, we have attempted to verify the robustness of CAPTCHAs offered by local e-commerce sites of Pakistan. We have successfully decoded the CAPTCHAs offered by these sites. Our proposed algorithms achieved promising results on all attacked CAPTCHAs. Using our thresholding methods, CFS, Recognition based segmentation methods and machine learning techniques we have successfully decoded the said CAPTCHAs with an overall precision of up to 82.4%.
关键词：CATPCHAS; HIPs; Color Filling Segmentation; thinning; CNNs.
Loading...
