【作者】 张肃；

【导师】 王文生；

【作者基本信息】 长春理工大学 ， 光学工程， 2014， 博士

【摘要】 应用光学相关原理的联合变换相关器以运算速度快、信息存储容量大以及平行运算等优点广泛应用于模式识别领域中。通过探测目标与模板而得到的相关点为依据,进而确定目标的方位信息,并将这种光电混合系统结合各种计算机编程技术,已经完成了对微光目标、红外目标、复杂背景下目标及小目标等多种静止目标的识别。但是在运动目标的光学相关识别中,由于目标自身运动所导致的目标与模板之间存在的畸变问题及周围复杂背景、天气变化和低对比度环境等外在因素对联合变换相关器的干扰,出现了在探测过程中没有相关点或相关点微弱的现象,严重影响运动目标光学相关识别的顺利进行。针对运动过程中目标与模板之间存在的大小、旋转及形状等不匹配的问题,提出了瞬态模板更换的方法,选择识别动态序列中的前一帧瞬时状态作为下一帧的模板,这样便可完成实时对模板的更新。为了增强相关点的亮度,提高相关器的识别效率,分别将小波、多小波及小波的提升算法引入对运动图像的物面处理技术中。提出了采用形态学膨胀处理的小波边缘提取算法及小波多尺度边缘融合算法；应用多小波所具有的多重多分辨率分析性质,提出了基于多小波变换的边缘提取算法及能分别增强高、低频信息的多小波图像增强技术；并进一步对小波的算法进行提升,给出了基于小波提升算法的边缘提取方法。分别将这些物面处理算法与瞬态模板方法相结合应用到光学相关运动目标的识别中,结果表明,基于小波变换的模极大值提取、小波多尺度边缘融合及基于多小波变换的图像增强算法均可提高相关器对低对比度运动目标的识别率。基于多小波变换的边缘提取可完成对运动微光目标及复杂背景下运动小目标的相关识别。而基于小波提升算法的边缘提取方法可实现复杂背景下运动目标的探测,并使运算速度得以提升,运算量减少。大量实验结果验证了应用该算法可实现光学相关运动目标的识别与探测技术。更多还原

【Abstract】 Joint transform correlator, applying the optical correlation theory, was widely used in the domian of pattern recognition for its merits of parallel, large capacity and high-speed. According to the correlation peaks between the target and the reference template the orientation information of the target can be determined. Applying this opto-electronic hybrid system with computer program technology has realized many stationary targets recognition, such as infrared target, low light level target, target in clutter scene, small target and so on.But in the course of the moving target recognition, there are many distortion problems betweem the target and the template because of the moving, and the external factors, such as the cluttered background, weather conditions and the low contrast environment, can also influence the joint transform correlator, so that there are no correlation peaks or low brightness of correlation peaks emerging, which seriously affects the optical correlation recognition of the moving target.To solve the size, rotation and shape mismatch problem between the target and the template, temporal template replacement method is proposed, which can take the temporal state of the target former frame as the template for the next frame to reach the purpose of real time template update.To enhance the brightness of the correlation peaks and improve the recognition efficiency of correlator, the wavelet, multi-wavelet and wavelet lifting algorithm are introduced in object plane processing technology of the dynamic sequences. The edge extracton based on wavelet combined with morphological dilation and wavelet multi-scale edge fusion algorithm are raised; Appling the multiple multi-resolution analysis property of the multi-wavelet, the edge extraction and image enhancement technology based on multi-wavelet are proposed; in addition, according to the wavelet lifting algorithm, the edge extraction method based on lifting wavelet is presented.Applying these object plane processing algorithms with temporal template method to optical correlation recognition of the moving target, the experiments show that edge extraction based on wavelet, wavelet multi-scale edge fusion and image enhancement based on multi-wavelet algorithm all can improve the the low contrast moving target recognition ratio of the correlator. Edge extraction based on multi-wavelet transform can achieve the correlation recognition of the moving low light level target and small moving target in the cluttered background. And the edge extraction based on wavelet lifting method can realize the moving target recognition in cluttered background, which can also improve the calculation speed and greatly reduce the amount of the computation. The large number of experiment results show that applying this algorithm can implement the optical correlation recognition and detection technology of the moving target successfully.更多还原