Ks
Ying Chen
WAVELET -- BASED IMAGE MATCHING for DIFFERENT SENSOR
Ying Chen
The Department of Surveying and Geo-Informatics
Tong Ji University, Shanghai China, 200092
KEY WORDS: Edge Feature, Wavelet, Detection, Line Moment, Matching.
ABSTRACT
In a general, different sensor images of the same scene have different gray characteristics. Simple algorithms such as
those based on area correlation can not be used directly. Therefore, an edge feature-matching algorithm is pressed in
this paper. The algorithm uses feature parameters of edges as matching primitives. An edge detection approach with
wavelet is discussed in order to extract edges from images. The mathematical models of edge representation and
matching algorithm based on line moments are developed. Results obtained from above methods are reported.
1 INTRODUCTION
In real-time matching for different sensor images during position and navigation of aerocraft, speed and precision are a
challenging problem. The method that combined feature with image least square matching is an effective approach. The
former may provide some credible initial values during the large-area searching procedure. The latter can realize exact
sub-pixel matching. In feature matching, reliability of feature extraction and detection is the one of key points. For these
we present an efficient detection algorithm used gradient direction profile.
3
In real-time scene matching, another problem is that it maybe exist bigger rotation between the real-time scene and
reference image. So we adopt the invariant-based line moment feature matching which is rotation-independent, and the
constraints are introduced. According to these, The sub-pixel matching technique that combined feature with least
square image matching is realized.
2 EDGE DETECTION BASED ON WAVELET TRANSFORM
In two-dimensional space, V (x,y) is said to be wavelet function if and only if it satisfies.
JJ Ce. yxy = 0
x
It is known that a two-dimensional smoothing function (x,y) whose integral is nonzero can be made as wavelet
function using its partial derivatives along x and y directions respectively.
ex,
Vin n
s (1)
y^ (x, y) 2208
2
For any function f(x, y) € L’(R*) in dyadic scale 2^ , its two-dimensional dyadic wavelet transform is defined as
the set of functions.
WE = QW f(x, y)W2 fo y)) je.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 177
