Research On Local Image Feature Extraction For Scene Perception

Image feature extraction is an important research content of image processing and computer vision.The local image feature is an important part of the image feature,which reflects the change of local information in the image,and is not susceptible to interference such as noise,light,scale,rotation and so on.In recent years,UAV aerial technology has leaped forward.However,the UAV aerial images have larger size and more complex image information,and they are susceptible to noise,light conversion,scale transformation,rotation and other issues of interference.Therefore,it is of great theoretical and practical value to study the feature extraction algorithm which can extract the features of UAV aerial images quickly and accurately,and obtain the features of high robustness.This thesis focused on Scale Invariant Feature Transform algorithm,to study a improved feature extraction algorithm for UAV aerial image.The main research work is as follows:(1)This thesis studied the basic theories of image feature extraction algorithm,and the two major classification of image features--global image features and local image features were studied and analyzed in this thesis.In addition,the thesis carried on the special research on the evaluation standard of the image feature extraction algorithm,including computational complexity of algorithm and repeatability of detectors.(2)This thesis focused on the theory research of local image feature extraction algorithm,especially the algorithm of Moravec,Harris,Smallest Univalue Segment Assimilatating Nucleus and Scale Invariant Feature Transform algorithm.Then experiments were carried out based on algorithms above.Combined with the theories of these algorithms,the intuitive evaluation of algorithms were obtained,showing that the local feature extraction algorithms have the superior performance in the technology of image processing.(3)Based on the Scale Invariant Feature Transform algorithm,combined with Harris corner detection algorithm and Smallest Univalue Segment Assimilatating Nucleus feature detection algorithm,a improved feature extraction algorithm was proposeed,which is more suitable for the feature extraction of UAV aerial images.Based on the magnanimity test data,compared with the Scale Invariant Feature Transform algorithm,the time complexity analysis and robustness analysis of the improved algorithm was concluded.The obustness analysis includes the noise immunity performance analysis,scale invariance analysis and rotation invariance analysis.(4)The basic model and algorithm platform of image processing and computer vision have been set up,and most of the algorithms in image feature extraction and description module have been written,debugged,transplanted and integrated in the algorithm platform.