Research On SURF Feature Extraction Algorithm Based On Visual Significance

In recent years,the research of image feature extraction has become more and more intensive in the field of computer vision and image processing.Feature extraction is not only a basic problem in image analysis,but also a key link from image processing to image understanding.It provides an important basis for the following steps of computer image processing: image registration and fusion,image classification and target recognition and image retrieval.Because of the influence of the shooting factors such as equipment,illumination,angle and so on,some fuzzy and reflective phenomena will appear in the collected images,which cannot be directly used for feature extraction.In this thesis,the image is preprocessed by mathematical morphology.Computer simulation of human vision,the information obtained from images is often very rich,which contains a large number of redundant information.To quickly and accurately obtain the key information we need,people have done a lot of research,Therefore,visual significance detection has been developed rapidly.In this thesis,SURF feature extraction methods were studied and improved in combination with the advantages of visual salience.The main contents are as follows:Firstly,this thesis briefly introduces the significance of image feature extraction,enumerates some classical feature extraction methods,and carries out simulation experiments on these algorithms.Secondly,this thesis introduces the basic theoretical knowledge and algorithm of mathematical morphology,makes simulation experiments and analyses them.Based on the effect of corrosion and expansion operation,a fast robust algorithm based on improved mathematical morphology is used to extract image feature points for images with high contrast.Firstly,the morphological algorithm is used to determine the range contour of the target in the image,which preserves the details of the image.On this basis,the final SURF feature points are extracted,which is more stable.The algorithm is applied to medical anterior segment images to provide effective feature points for doctors to diagnose the disease.Then,this thesis introduced in detail the theoretical knowledge and algorithms of SURF feature extraction and visual significance detection,and analyzed the experimental results.Based on the experimental results of the FT visual saliency algorithm,an image-guided filter is used for the filtering part of the image.Comparedwith the original Gaussian filter,the mean and variance of the pixels in the image domain can be used as local estimation,so that according to the image The information is adaptively adjusted to adjust the output image weights to get better results.Secondly,SURF algorithm was improved.In the fourth stage of SURF algorithm,improved FT algorithm and feature point weight algorithm were used to eliminate feature points that were not important in structural information,so as to obtain more ideal feature points.Finally,the algorithm is applied to the general scene graph,and the results show that the proposed algorithm can ensure the efficiency of feature points while extracting not many feature points,and maintain a high matching rate while matching the image feature detection,which is helpful for the continued processing of subsequent images.The algorithm is applied to medical anterior segment images to extract significant lesion feature points,which can provide an effective reference for ophthalmologists to diagnose the disease.