the method ASIFT
the comparison of images
V.V. Voronin, V.I. Marchuk, M.M. Pismenskova, T.V. Morozova
Currently, computer vision is a particular stage of development. This area has become popular due to the occurrence of many multimedia applications using computer vision and ideas of mass deployment in modern life means of processing visual data. However, despite the rapid rate of development, there remain a number of tasks to be quite problematic. These include the problem of recognition and matching images. This is due to the lack of imaginative perception of visual information computer. Despite the ease with which people solve this problem, there is no universal approach that allows to develop an algorithm of object recognition computer. This task is urgent and the solution potentially useful in many fields, ranging from motion detection, automatic system monitoring, management, robotics and ending search for duplicate images in a database, based on the search for matching images.
One of the main approaches to image search is a search on local descriptors. Local descriptors are a feature vector constructed from the individual fragments of images. That is, they do not cover the whole image, and contain only information about the selected fragments in some way.
We propose a modification of the method of searching and comparing images ASIFT based detection and comparison of point features. The presented modification allows you to define tags on the mirror surfaces of images. Research shows the effectiveness of the modified method of searching on the basis of local descriptors based on the test images, in comparison with the method ASIFT, which showed the effectiveness of the proposed method.