Research And Design Of Image Feature Extraction Algorithm Based On Local Information

With the rapid development of information technique,the amount of image data and image category is increasing heavily.It is becoming a more and more important task that analyzing and understanding images automatically.As a basic step towards image matching,image classification and image understanding,feature extraction is palying an extremely important role in computer vision such as image classification,image matching and image retrieval.Due to the fact that the number of pictures is very big and one picture may carry different information in different envoriment.It is a very difficult task to extract image features in an effective way in the real world.Among all image features,local feature has a better performance in coping with noise interference,the change of angle and image rotating.Threrfore,the local feature is widely researched in image processing.This dissertation introduces the background and current research situation of feature extraction and gives a detailed introduction of SIFT algorithm.Several classical image feature extraction algorithms are analyzed and the detailed theory of these algorithms is introduced.Then,the advantage and disadvantage of these algorithms is discussed.Based on SIFT,three improved algorithms are proposed in this dissertation.The main content of this thesis is as follows:(1)For the SIFT algorithm,since the feature points of background has a negative effect on feature extracting and matching rate,this thesis gives an improved feature extraction algorithm based on adaptive canny edge detection.At the beginning,using canny algorithm to detect the edge and remove some background pixel points.Then the SIFT algorithm is used to get feature points in the neighbourhood of the edge.The proposed algorithm removes the feature points from the background,which decreases the effect of the background.Simulation results show that the new algorithm has a better performance in image matching.(2)In order to decrease the complexity of the feature extraction algorithm,an adaptive pixel step size algorithm is used to decrease the number of the feature points got in the Gaussian Pyramid.The simulation results show that the algorithm performs well in reducing the time-consuming.Although the matching rate reduces but the algorithm has a good performance when the picture is complicated and a high real-time performance is demanded.(3)The matching rate has an obviously decrease when a picture contains several similar parts.This is a limitation of SIFT algorithm.Contourlet transformation is applied.These key points of the new algorithm is that it can introduce global texture information to the SIFT local features.Thus the robustness of the improved algorithm is guaranteed.Simulation results show that the new algorithm has a good performance.