Research On Methods For Infrared Image Target Segmentation

Infrared guidance technology is the key development direction of precision–guided weapons. The object detection, segmentation and recognition are the main technologies of infrared guidance system under the condition of infrared imaging. As the foundation of object recognition and tracking, infrared image segmentation plays an important role in improving the performance of infrared pre-warning and guidance systems.The main task of this thesis is to research the segmentation algorithms for infrared image objects under complex background. Firstly, since the traditional thresholding segmentation methods have a bad performance in the case of complex background or the object gray distribution which is asymmetric, an improved segmentation algorithm is proposed based on the isoperimetric ratio by analyzing the variation of isoperimetric ratio with gray levels. The proposed method performs very well, in the case of the object can be separated from the background. Secondly, for the thresholding segmentation methods are effectively just under some special condition, the segmentation algorithm is studied based on mean shift with multi-dimension characters for infrared image object. However, the traditional mean shift method is time-consuming, a new fast mean shift algorithm is introduced to decrease the complexity of mean shift by improving the traditional search manner. On the base of fast mean shift filter, a new segmentation method of gray relation analysis for infrared image object is put forward to solve the over-segmentation problem produced by the mean shift and separate object from complex background accurately. Finally, for the classical evaluation measures are ineffective to evaluate the performance of segmentation algorithms, five character indexes are selected to construct a synthesis goodness measure. Experiment results show that the new measure can evaluate segmentation algorithms effectively.