2004
Rmse
Rmse
8,66
7,89
10,03
Rmse
0,10
0,11
—r
Rmse
——
13.70
13,03
——
13,61
OBJECT RECOGNITION BASED ON
TEMPLATE CORRELATION IN REMOTE SENSING IMAGE
Rhe Yi 7
Jun Zhang^^*, Xiyuan Zhou"
; m Dept. E.E., Beijing Institute of Technology, Beijing 100081
Communication Telemetry & Telecontrol Research Institute, Shijiazhuang 050081
junzhang@fescomail.net
KEY WORDS: Correlation, Matching, Extraction, High resolution, Object Recognition
ABSTRACT:
In recent years, the spatial resolution of remote sensing image becomes much more higher then ten years ago. There are more
information reflected by modern remote sensing image. The research of image processing and analyzing based on traditional low
resolution image has already not satisfied the need for people to get more accuracy information from high resolution remote sensing
image. People want to get information about some particular objects and the change about a particular area from remote sensing
image, this is particularly important to the urban plan and disaster surveillance. On the base of analysis of the conventional methods
for information extracting from the remote sensing image, a method of extraction particular object in remote sensing image based on
feature template correlation is proposed. The method includes three parts: building the template, image match and template
correlations, and object recognition. The methods are applied to several high-resolution example images, and vehicles as example
object in the image are extracted and recognized. Those examples illuminate that the method proposed in this paper is effective and
accuracy.
1. INTRODUCTION
With the increasing improvement of spatial resolution of remote
sensing image, we can acquire more information about object
on the earth. Based on conventional remote sensing image
processing method, we can classify different type of large
terrain, such as city and farmland. When the resolution of
remote sensing image approaches to 1 meter or even less, we
can see most small objects on the ground clearly, such as
houses, vehicles, and so on. It is difficult to distinguish those
small objects from image background by conventional remote
sensing image processing methods. ^ Now there are many
studies on man made object (roads, houses, vehicles)
recognition in the high-resolution image (Rucklidge, 1997;
Rensheng, 1997; Ballard, 1981; Selvarajan, 2001).
Object recognition algorithm in optical camera image
processing are applied to the remote sensing image because the
improvement of the spatial resolution of image. but there are
several different problem faced to remote sensing image
processing and optical camera image processing, 1) the spatial
resolution of remote sensing image is relatively low although
much improved; 2) the remote sensing image are acquired with
different viewpoint angle and view field; 3) the SNR of the
image is relatively low; 4) the object in remote sensing image
usually has scale, translation and distortion; 5) the ratio of the
number of object pixel and whole image pixel is quite small.
Because the object on the earth are quite variety, from a large
city to a small vehicle, and one object has different appearance
on the remote sensing image because of different view point,
view field, view angle and different climate condition. The
algorithm based on supervised or unsupervised methods have
good performance to classify large terrain objects. But for small
objects in the remote sensing image, the algorithm for
recognition should be paid more attention to study.
In this paper, to get accuracy classification of objects,
hierarchical object template database should be built. A wavelet
transform and morphological processing is used to extract the
feature of image and to find the interested object region.
Second, template-matching methods are discussed based shape
feature template, template-correlation is a time-consuming, the
template and the image are never the same, so the method for
correlation should be robust to matching noise and time —saving.
887
Section 2 gives profile of the approach and section 3,4,5 present
the detail description; Section 6 concludes the object
recognition processing.
2. OUTLINE OF THE APPROACH
In this section a coarse description of the algorithm is presented.
The whole process is summarized in the flowchart of Figure 1.
In section 3, the single steps are explained in detail.
Feature extraction
Y
Hierarchical
Template database
ROI extraction
ROI windows
Y
Common
Feature extraction
Object first-level
Feature template
|
Object second-level
Feature template
Object candidates
Object last-level
Feature template
Object candidates
Li Object recognition
Figure 1. Flowchart of the algorithm that is used to recognize
the object in the remote sensing image based on hierarchical
template ,
