چكيده به لاتين
The purpose of this dissertation is present of robust and highly reliable small dim aerial target detection methods based on human visual attention system in infrared images with complicated background clutter. Detection of incoming small targets from a long distance is of paramount importance in defence systems. Therefore, targets in the acquired IR images are usually dim and small while clutter and noise in complex background are high, make the detection of the small targets with high reliability and low false alarm rate challenging.
Small target in IR images is as salient features that attracts human visual attention. Nowadays, developed computational models based on human visual system are proposed to be used in many applications including target detection in IR images that have been desired results. Hence, in this thesis in addition to investigating common methods for small target detection in IR images and related basic concepts and problems, human visual system and proposed computational models related to small IR target detection investigated and analyzed.
We propose two robust and highly reliable methods based on human visual attention system so they can detect small dim target in IR images with complicated clutter effectively. In the first method, by using the static and motion saliency maps fusion, emphasizing the obtained saliencies from one method to another and applying the information and benefits of all maps in the saliency map fusion, this method suppresses the background clutter and noise with high reliability, makes the target more prominent and finally increases the contrast between them. Therefore, the target could be easily detected in the fused saliency map.
In the second proposed method, first, the Gaussian-like feature maps are extracted from the original image. Then, saliency maps are created based on Pulsed discrete Cosine Transform (PCT) in which the target is salient and background clutter is suppressed. Finally, to increase the contrast between the target and background clutter and to raise robustness of this method against false alarms, saliency maps are fused adaptively and target detection is accomplished using it.
To evaluate the performance of the proposed methods, two data sets of real-life IR images containing moving small target in complex clutter have been prepared in collaboration with Imam Hussein University. The experiments are carried out on the data sets and qualitative and quantitative assessments show that the proposed methods can detect small targets in IR image with high reliability and are more effective compared with other methods based on human visual attention. So that in one of the proposed methods, quantitative measure of signal to clutter ratio has an average rate of 49%, which is 33% higher than the most robust compared method. This difference indicated that the proposed method can suppress background clutter with high reliability regarding to other methods, while at the same time points to maintaining the target signal. Therefore, the proposed methods can be used in many applications for detection of small targets in the IR image
