Technology Research And Application Of Image Matching Based On Feature Points

Image matching is a process to find corresponding points between two images or aplurality of images by a matching algorithm. As an important technology in the field of imageinformation processing, Image matching has been widely used in machine vision filed,moving object detection field, medical analysis field, industrial inspection field, etc. Andthe images used to matching are usually images with the same object taken in different time,different angles, different locations or different weather conditions by camera. In specificapplications of image matching, it often occurs in among the images with differentcomplexities or a large number of images, therefore, our goal is to find a matching algorithmwhich own high matching precision rate and well real-time performance. The current imagematching algorithms include the image matching method based on image gray and the imagematching method based on image features. The first method is easy to achieve and grasp, andhave good matching accuracy, but does not have rotational invariance, scaling invariance andscaling resistance; the second method is suitable for most areas, but have large amount ofcalculation and the method is achieve complicated. This paper focus on the image matchingmethod based on image features, and main study methods of feature extraction and improvethe existed feature extraction method in order to achieve an algorithm with high accuracy andgood real-time. The main contents of this paper are:⑴Introduce the basic methods of image matching, and deeply study the imagematching methods based on image feature, and learn the theory of classical methods aboutlocal feature detection, and analysis the performance on classical extraction featuremethods.⑵Use the SIFT algorithm to extract the trademark image feature exists limitations. Aimat the existing limitations, we proposed an improved method which combined the SIFTfeature and the Harris feature, and replace the Harris feature describe method by the SIFTfeature described method. The improved method makes use of corner point detection theadvantages of low amount calculation, less time, and the feature uniform distribution. Theexperiments proved that, the improved method based on the SIFT feature and the Harrisfeature extract the features of trademark image have the advantage of high efficiency andreal-time characteristics. ⑶The classical SIFT feature extraction algorithm, can used to extract the colorfulimages with large affine angle features, and can get a large amount of feature points, but needlong time; While use SURF algorithm to extract the colorful images with large affine anglefeatures, have high mistake matching rate and loss the colorful information. Thus, weproposed a method of the affine invariance for color image registration, which obtained aseries of analog images by simulate the different affine angles, then calculate the colorinvariants, and extract features and match the feature points. The experiments proved that theproposed method extracts feature points need shorter time and have higher match rate, canwell solve the mismatch problem of the color image caused by the large affine angle.