Date of Award


Document Type

Thesis (Master's)

Department or Program

Department of Computer Science

First Advisor

Sean Smith


Maintaining usable security in application domains such as healthcare or power systems requires an ongoing conversation among stakeholders such as end-users, administrators, developers, and policy makers. Each party has power to influence the design and implementation of the application and its security posture, and effective communication among stakeholders is one key to achieving influence and adapting an application to meet evolving needs. In this thesis, we develop a system that combines keyboard/video/mouse (KVM) capture with automatic text redaction to produce precise technical content that can enrich stakeholder communications, improve end-user influence on system evolution, and help reveal the definition of ``usable security.'' Text-redacted screen captures reduce sensitivity of captured material and thus can facilitate timely data sharing among stakeholders. KVM-based capture makes our system both application and operating-system independent because it eliminates software-interface dependencies on capture targets. Thus, our work can be used to instrument closed or certified systems where capture software cannot be installed or documentation and support lack. It can instrument widely-varying platforms that lack standards-compliance and interoperability or redact special document formats while displayed onscreen. We present three techniques for redacting text from screenshots and two redaction applications. One application can capture, text redact, and edit screen video and the other can text redact and edit static screenshots. We also present empirical measurements of redaction effectiveness and processing latency to demonstrate system performance. When applied to our principal dataset, redaction removes text with over 93\% accuracy and simultaneously preserves more than 76\% of image pixels on average. Thus by default, it retains more visual context than a technique such as blindly redacting entire screenshots. Finally, our system redacts each screenshot in 0.1 to 21 seconds depending on which technique it applies.


Originally posted in the Dartmouth College Computer Science Technical Report Series, number TR2011-690.