Millimeter wave imaging technology provides a new detection method for security, fast and
safe. But the wave of the images is its own shortcomings, such as noise and low sensitivity.
Systems used for security, since only the corresponding specific objects to retain the information,
and other information missing, so the actual image is difficult to locate in the millimeter wave .
Image fusion approach can be used to effectively solve this problem. People usually use visible
and millimeter-wave image fusion. The use of visible image contains the visual information. The
fused image can be more convenient site for the detection of concealed weapons and to provide
accurate positioning. The integration of information from different detectors, and there are
different between the two levels of signal to noise ratio and pixel resolution, so traditional
pixel-level fusion methods often cannot satisfy the fusion. Many experts and scholars apply
wavelet transform approach to deal with some remote sensing image fusion, and the performance
has been greatly improved. Due to these wavelet transform algorithm with complexity and large
amount of computation, many algorithms are still in research stage.
In order to improve the fusion performance and gain the real-time image fusion, an Integer
Wavelet Transform CDF97 based with regional energy enhancement fusion algorithm is proposed
in this paper. First, this paper studies of choice of wavelet operator. The paper invites several
characteristics to evaluate the performance of wavelet operator used in image fusion. Results show
that CDF97 wavelet fusion performance is better than traditional wavelet wavelets such as db
wavelet, the vanishing moment longer the better. CDF97 wavelet has good energy concentration
characteristic. The low frequency region of the transformed image contains almost the whole
The target in millimeter wave image often has the low-pass characteristics and with a higher
energy compare to the ambient region. Based on this assumption, a new fusion rule is proposed here. Firstly, get the low-frequency part of the selection matrix according to the comparison
regional low-frequency energy matrix of the two images. Then, taking into account the
consistency and continuity of low-frequency detail, high frequency and low frequency selection
matrix can be the same. And this can ensure a better fusion of the edge of the image
characteristics. Simulation results show that this algorithm's performance is much better compared
to traditional energy-weighted and average method, though less quantitative comparison point
compared to the region based method, yet, the difference is small according to the visible
This algorithm is tested in the self-developed image processing platform based on DM642. By
using the optimization strategy, the speed of 256 × 256 dual-channel image fusion can be more
than 28 F/S. Therefore, the proposed fusion algorithm can meet the system performance and
Automatic target detection (ATD) in infrared imagery is challenging problem to Surveillance and Reconnaissance
application. Though the recent of advances of remote sensing instrument which significantly improve system's spatial
resolution and sensitivity, the size of many infrared targets is equivalent to or even smaller than pixel resolution.
Therefore target recognition must be carried out within one pixel. Under such circumstance, traditional image processing
techniques generally can not perform well in applications.
The paper presented here investigates the issue and presents a novel algorithm for automatic target recognition in dual
band infrared imagery with no priori knowledge. It consists of two stage processes, clutter rejection in infrared imagery
of each band and followed by target detection with the fusion of the two bands of infrared images. Morphologic method
instead of spatial-based processing techniques is applied in clutter rejection. An alternative way is Top-hat which gains a
good performance in images with small targets. An in fusion center, dual-band infrared images are fused according to the
AND rule whereas the decision rules are obtained from Neyman-Pearson (NP) criterion. A simple method of gaining
each sensor's threshold with optimization has been developed.
Relative simulations, assuming noise in infrared system is subjected to Gaussian distribution, shows the fusion result
with a greater performance than either one even one's performance has been greatly degraded. Experimental results are
presented using dual bands of registered infrared imagery to evaluate the performance of the new algorithm. It is
suggested that the method presented above is robust, feasible and effective in this application.