Attribute Reduction Of SIFT Algorithm Local Descriptor Based On PCA And Class Attribute Interdependence Minimization

Image feature retrieval, representation, matching and recognition is thehot research contents of virtual reality, computer vision and other relate fields, extractionof stable, efficient image features of digital image processing technique is the mostimportant part of the foundation. Along with the development of image technology,computer technology and video processing technology, especially with the rapiddevelopment of video processing technology, higher requirement of real-time in imagematching and recognition be required.Based on the image feature matching technology, which is the most important task isto extract stable image features and describe it. The most commonly used methods iswavelet algorithm, the algorithm based on spatial relation, invariant descriptor algorithm,SIFT algorithm and so on. Now The scale invariant feature transform (SIFT-ScaleInvariant Feature Transform) feature matching algorithm is a most popular algorithm indomestic and foreign, and also is good in matching result performance, the algorithmusing multi-scale space superiority, to achieve feature retrieval and matching underdifferent scale. Based on multi-scale image feature matching technique can handle withthe different characteristics between the image and the image of deformation of complexcases, can be stably for two displacement relatively large images for accurate featuredetection and matching. So SIFT algorithm matching capability is stronger, can extractmore stable image feature, can handle the image when affine transformation, illumination,rotation transform, perspective transform, and also has a relatively stable image featurematching ability to some extent arbitrary angle acquisition image, therefore can handle therelatively large differences between the two images to match the characteristics of theproblem. But the SIFT descriptor has some obvious deficiencies what are dimensions high,more redundant information, real-time poor when matching and recognition.Principal component analysis (PCA-Principal Component Analysis) method is themost widely used feature extraction and data reduction method. It can be provided basedon minimizing the reconstruction error optimization feature extraction and reduction, butthe method of principal component analysis is the interception of data collection features,so lead to loss of information, also did not consider the characteristic attributes of thespecial and typical. Moreover, according to the characteristics of the physical meaning ofthe vector, using principal component analysis method will still have the data redundancy.Class attribute dependent minimum (CAIM-Class Attribute InterdependenceMinimization) algorithm is to study the characteristic pattern of characteristiccomponents and characteristics of interdependence, through statistical and screening, andthen retention the best embody the characteristics of special and typical features, combined principal component analysis with class attribute dependence of minimummethod can effectively remove the feature data redundancy, to improve the speed andaccuracy of image matching.In this paper, combined principal component analysis method with the class attributedependence of minimum method on SIFT feature descriptor dimension reduction, theexperimental results show that the algorithm for image feature matching of the matchingspeed is improved obviously, correct match rate also has a certain improvement. It showsthat this method can extract effective and stable image features.