In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
classes of interest at the two dates against all those identified as
changed pixels may provide some degree of information on the
type of land-cover transition, but not with the accuracy and
reliability provided by fully-supervised approaches).
In this paper, we formulate this complex issue in terms of a
compound decision problem (Duda et al., 2000) and propose a
novel partially-supervised change-detection (PSCD) technique
capable of exploiting the only prior knowledge available for the
specific land-cover classes of interest at the two times (thus
avoiding the need to rely on exhaustive ground-truth informa
tion for all the classes), while providing accuracies comparable
with those of fully-supervised methods. Moreover, it has the
great advantage of being sensor-independent, which allows
selecting at each date the set of sensors and features most suit
able for characterizing the targeted class(es) of interest.
The proposed method aims at estimating at each date the prob
ability density function (PDF) and the prior of both the class(es)
of interest and the remaining unknown land-cover classes (for
which no ground truth is available) represented as a single un
known information class. In particular, PDFs are approximated
by a mixture of suitable basis functions whose free parameters
are determined employing the iterative Expectation-
Maximization (EM) algorithm (Dempster et al., 1977). Changed
pixels are then identified using an iterative labelling strategy
based on Markov random fields (MRF) (Solberg et al., 1996)
which allows taking into account both spatial and temporal
correlation between the two images, as well as properly con
straining the probability estimates.
For demonstrating the capabilities of the proposed method,
extensive experimental trials have been carried out with differ
ent combinations of multispectral, hyperspectral and SAR data.
Obtained results confirmed the effectiveness and the reliability
of the proposed technique, which provided very promising re
sults. In particular, accuracies are comparable to those achieved
with the PCC method in the presence of an exhaustive ground
truth for each image both considering the maximum likelihood
(ML) classifier (Richards and Jia, 2006), as well as support
vector machines (SYM) (Cristianini and Shawe-Taylor, 2000).
2. PROBLEM FORMULATION AND ASSUMPTIONS
For the sake of simplicity we will describe the problem and the
proposed method under the assumption of a single targeted
land-cover transition of interest. The extension to the case of
multiple transitions is straightforward.
Let us consider two IxJ co-registered remote-sensing images
*'={*#}&, and X 2 = {xl}';U, 4 e R 02 , referring
the same geographical area at times t\ and h , respectively,
where x!y, xj represent corresponding feature vectors (even
derived from different sets of sensors at each date, respectively,
and merged using a stacked vector approach (Richards and Jia,
2006)) associated with the pixel at position (i,j) , and D\, Eh.
define respective dimensionalities.
Let £1'={cd ,...,co.L} and Q 2 ={ed ,...,ceL) be the set of land-
cover classes characterizing X' and X 1 , respectively. In the
following, we will denote as cdt e £2' and cdt e Q 2 the infor
mation classes of interest at t\ and h , for which Ni and N2
labelled training patters are available, respectively. Elence,
coU ={Q‘-ceD and aLk - {Q. 1 -cd*.} will represent corre
sponding unknown classes (each consisting of the merger of
remaining classes, for which no ground truth is available).
Let C and C 2 = {Cy }';/=, denote two sets of labels for
X' and X 2 , respectively, where Cle {cd M ,coU} and
C|e {ainx,ocLi} are associated with the pixel at position (i,j).
In this framework, our aim is to identify the two sets C, C 2
maximizing the posterior probability given the two images X 1
X 2 and, finally, to draw pixels experiencing the targeted land-
cover transition from edit to cdt. This can be formalized as a
compound decision problem (Duda et al., 2000):
{C',C 2 }=argmax{P(C l ,C 2 \X\X 2 )} (i)
c‘,c 2
According with the Bayes theory, finding a solution to (1) is
equal to determine the sets of labels maximizing the likelihood
C(X',X 2 \C',C 2 ) = P(C',C 2 )-p(X',X 2 \C',C 2 ).
In the reasonable hypothesis of time-conditional independence,
the problem can be written as:
{C',C 2 }=argmax{C{X ,?d \ C ,C 2 )-P(C',C 2 )p(X\C')-p(X 2 \C 2 )) (2)
where p(X'\C') and p(X 2 \C 2 ) represent the conditional
PDFs at t\ and ti respectively.
3. PROPOSED PARTIALLY-SUPERVISED CHANGE-
DETECTION TECHNIQUE
For addressing the complex task described in Section 2, we
propose a novel partially-supervised technique aimed at ap
proximating the class-conditional densities p(X' \C'),
p(X 2 \C 2 ) as mixtures of suitable basis kernel functions and
estimating the joint prior probability P(C',C 2 ) properly taking
into consideration the spatio-temporal context.
The rationale is based on the observation that the PDF of an
image can be always approximated by a mixture of suitable
kernels (i.e., Parzen density estimation (Duda et al., 2000)).
Accordingly, similarly to what is commonly done in the context
of Radial Basis Function Neural Networks (RBF-NN) (Bruz-
zone and Femandez-Prieto, 1999), we model for each pixel of
both images the PDFs p(x\), p(xl) as a mixture of K circu
larly symmetric multivariate Gaussian functions. Kernel pa
rameters (i.e., centres and variances) are initialized using the k-
mean clustering algorithm (Bruzzone and Femandez-Prieto,
1999), whereas final estimates are obtained by using the EM
algorithm (Dempster et al., 1977). Then, class-conditional den
sities of the interest class p(x)j\cdt), p(x 2 j \ cdt) are modelled
by properly weighting the resulting set of kernels using again
the EM algorithm over the training samples available for (dt
and cdt. This is somewhat analogous to the training phase of
RBF-NN which is generally carried out in two steps: i) selec
tion of centres and variances of the kernel functions associated
with hidden units on the basis of clustering techniques; and ii)
computation of weights associated with the connections be
tween the hidden and output layers on the basis of available
training patterns.
The PDF of the entire image is itself a mixture of the interest
and unknown class-conditional densities, weighted by corre
sponding prior probabilities. Accordingly, we obtain a first
rough approximation for p(x\ |coU), p{x] \ ednk) initializing
priors to 0.5. Afterwards, estimates are refined using a novel
MRF-based iterative labelling strategy accounting for spatio-
temporal correlation, which permits to model P(C!/,Cij) mutu
ally considering the local neighbourhood of each pixel in the
two images. Finally, changed pixels are identified and associ
ated with the targeted land-cover transition by minimizing a