Analysis and application of sports video image based on feature matching

In order to improve the accuracy and speed of sports video recognition, a sports video recognition model based on feature screening and support vector machine is proposed. Aiming at the disadvantages of the content-based image classification and retrieval method, such as huge amount of data and high computational complexity, this paper proposes an image classification method based on SIFT algorithm and applies it to sports image classification. On the basis of sorting image pixels, the extreme value region based on the change of gray closed value is extracted according to the neighborhood quadtree data structure optimized by rank merging and path compression, which effectively restores all the information of the region that eventually becomes a pixel value and gray threshold. The extreme value region is used as a node to construct a component tree, and the maximum stability criterion is obtained. In order to facilitate the subsequent feature description, a vector-based second-order center-rectangle general formula is constructed, and the general formula is reduced to a two-dimensional covariance matrix, and the irregular shape region is adjusted to an ellipse. Due to the complex and changeable image data, no matter what kind of algorithm is currently used, it cannot be applied to the matching problem of various images, especially for the situation that the image has scale change and affine deformation. Taking the image classification method in the SIFT algorithm as the research perspective, the characteristics of the SIFT algorithm and its generation are deeply analyzed, and the method is applied to the classification of sports images.