Akademik Veri Yönetim Sistemi
European Biotechnology Congress 2018, Athens, Yunanistan, 26 - 28 Nisan 2018, cilt.280, ss.24
Colonoscopy and wireless capsule endoscopy are the most common techniques
to monitor and detect abnormalities in the colon. During this process, probe or
capsule movement causes blurry images. Detection and removal of blurry images
is critical for further automatic abnormality detection procedures. Several
methods based on wavelet transform, Canny edge detection, discrete Fourier and
cosine transform have been proposed so far. The Laplacian operator-based
approaches have not been used on colon images yet. In this study, we extracted
four features from colon images based on Laplacian operator for the
discrimination of blurry images from visually normal images. The features were
the energy and variance of the Laplacian of the images, the average of pixels
obtained using diagonal and modified Laplacian operator. We used 80 frames (40
of them were blurry) selected from videos that are available in an open-source
database (https://www.gastrointestinalatlas.com/). The features were utilized
as the inputs to various classification methods, where cubic support vector
machines resulted in the best performance. The classification accuracy values
we obtained were 82.5%. The results of this study indicated that for the
automatic detection of blurry images in colon videos Laplacian operator-based
features were feasible.