Front. Comput. Sci., 2015, 9(5): 720–728
DOI 10.1007/s11704-015-4391-8
Adaptive segmentation based on multi-classification model for
dermoscopy images
Fengying XIE
1
, Yefen WU
1
,YangLI
1
, Zhiguo JIANG
1
,RusongMENG
2
1 Image Processing Center, School of Astronautics, Beihang University, Beijing 100191, China
2 General Hospital of the Air Force, People’s Liberation Army, Beijing 100036, China
c
Higher Education Press and Springer-Verlag Berlin Heidelberg 2015
Abstract Segmentation accuracy of dermoscopy images is
important in the computer-aided diagnosis of skin cancer and
a wide variety of segmentation methods for dermoscopy im-
ages have been developed. Considering that each method has
its strengths and weaknesses, a novel adaptive segmentation
framework based on multi-classification model is proposed
for dermoscopy images. Firstly, five patterns of images are
summarized according to the factors influencing segmenta-
tion. Then the matching relation is established between each
image pattern and its optimal segmentationmethod. Next, the
given image is classified into one of the five patterns by the
multi-classification model based on BP neural network. Fi-
nally, the optimal segmentation method for this image is se-
lected according to the matching relation, and then the image
is effectively segmented. Experiments show that the proposed
method delivers better accuracy and more robust segmenta-
tion results compared with the other seven state-of-the-art
methods.
Keywords adaptive segmentation, feature extraction, pat-
tern classification, dermoscopy image
1 Introduction
Skin cancer is one of the most rapidly increasing cancers in
the world, with an estimated annual incidence of 81 220 and
12 980 deaths in the United States in 2014 [1]. Although the
Received September 3, 2014; accepted January 21, 2015
E-mail: xfy_73@buaa.edu.cn
incidence of skin cancer in China is lower than the US, Eu-
rope and Australia, it has increased 3%–8% annually and
has doubled over the past decade. The earlier the malig-
nant melanoma (MM) is recognized, the higher probability
the melanoma is cured in many cases [2]. Dermoscopy, a
non-invasive skin imaging techniquewhichmakes subsurface
structures more easily visible, is introduced to improve the
accuracy in the diagnosis of pigmented skin lesions (PSLs)
[3].
Accurate segmentation for dermoscopy images is crucial
for the automatic diagnoses of skin cancer and it is one of the
most active areas in the computerized analysis of PSLs. It is
reported that 28% of the computer vision articles dating from
1984 to 2012 on “single lesion analysis” is concentrated on
“border detection” [4]. So many segmentation methods have
been proposed and developed but efforts are still made to get
more accurate border. The segmentation of lesion has been
challenged because the dermoscopy images are complicated
for [5]: i) low contrast between the lesion and the healthy
skin; ii) irregular and fuzzy lesion borders; iii) artifacts aris-
ing in the image such as hairs, and air bubbles; iv) variegated
color; v) fragmentation made by depigmentation. Generally,
most of the automatic methods can deliver satisfactory seg-
mentation for the images with strong contrast and homoge-
neous texture but suffer from the challenging cases more or
less. So taking the complicated and challenging cases of der-
moscopy images into consideration, it is difficult to find an
effective segmentation algorithm which is applicable for all
the dermoscopy images.
While for a complicated image set, if the most appropri-