31-07-2012, 03:44 PM
The Failure of Noise-Based Non-Continuous Audio Captch
The Failure of Noise-Based Non-Continuous Audio Captchas.pdf (Size: 1.54 MB / Downloads: 34)
INTRODUCTION
Many websites rely on Completely Automated Public
Turing tests to tell Computers and Humans Apart
(CAPTCHA1) [18] to limit abuse in online services such as
account registration. These tests distinguish between humans
and automated processes by presenting the user with a task
that is easy for humans but hard for computers. Designing
such tests, however, is becoming increasingly difficult
because of advances in machine learning. In particular, the
widely used category of image based captchas have received
close scrutiny recently [17], [25], [30], [31].
BACKGROUND
Decaptcha can be split into three main components: a segmentation
stage that extracts spoken digits, a representation
scheme for the extracted digits, and a classification stage
that recognizes each digit. While Decaptcha uses two active
phases, the intermediate representation is an important part
of a two-phase solver because of its impact on performance.
Accordingly, the first three subsections of this section provide
a high-level overview of the concepts used in each of the
three components. We then discuss the metrics that are used
to measure Decaptcha’s performance.
Regularized Least Squares Classification (RLSC).
RLSC
is an adaptation of a regression algorithm to binary classification
[22]. It was introduced as a fast yet similarly performing
alternative to the Support Vector Machine (SVM), one of
the most successful binary classifiers. In its simplest form,
RLSC finds a hyperplane w ∈ Rd that separates its training
data into positive and negative classes by solving