Research On Image Local Feature Description Algorithm Based On Intensity Order Pattern

With the development of computer technology,the invariant local feature of image becomes the research hotspot in the field of image processing,which is one of the basic problems in computer vision research.The difficulty of image local feature matching algorithm is how to construct the feature descriptor with invariance and strong robustness,which is also one of the key problems.Image local feature description is the basic step of many visual applications such as 3d reconstruction,image stitching and object recognition.Therefore,the image local feature description algorithm has been widely concerned by computer vision academia and industry for a long time.Usually,due to the diversity of image content and access way,the differences in image imaging conditions and scenes lead to the complexity of geometry,illumination,angle of view and image quality,Image local feature description algorithm research still faces many problems and challenges.The key difficulty is to extract the robust invariant feature descriptors with high recognition ability.In this paper,the local feature description algorithm based on intensity order is researched in detail,including the following aspects:1)Analyzed the feature description algorithm of local intensity order pattern,the algorithm constructs features by counting the intensity order patterns of the sampling points in the patches,and according to the difference information of the sampling points to calculate the value weight of each mode when constructing the feature descriptor.After analysis,the pattern of intensity order of sampling points is redundant,and only used the difference information between sampling points when calculating weight.For the above two problems,this paper proposes an improved method.Firstly,the pattern of low contribution and low participation in intensity order pattern was eliminated.Secondly,the weight of the corresponding pattern is calculated by combining the sequence structure and difference information of the sampling points,and the local intensity information is fully used.The experimental results show that the improved feature descriptor by the proposed method has good robustness and improves the description precision and identification effectively.2)An improved algorithm is proposed to solve the problem of overall intensity order pattern algorithm that the global information utilization is not sufficient and the feature dimension increases exponentially with the number of sampling points when constructing the feature descriptor.The proposed algorithm divides the feature descriptors into two parts,and the overall intensity order feature descriptor is formed by using them,so that the global information can be greatly utilized without adding the feature dimension while increasing the number of sampling points.3)The improved LIOP and OIOP features are combined which are complementary,and the PCA algorithm is used to obtain the improved MIOP feature descriptor.Experiments on standard Oxford data set,the results show that the proposed improved algorithm of feature descriptor has good robustness.The accuracy and identification of the description are improved,and the performance evaluation is better than other comparison algorithms.